Model rankings

Models ranked by quality, speed, and cost.

Compare the live MindsHub catalog by capability, response time, and measured cost per task. For models outside the catalog, see the blog.

Get your API key API docs

Use cases

What are you building?

Start with the free model. When a task needs more, pick by the kind of work you run.

Start here, free

Start free with the MindsHub Air model.

An alias we keep pointed at a balanced, general-purpose model, so the model behind it can change without a code change on your side. Free within fair use in MindsHub Cowork, Code and the API. Call it as mindshub_air.

When a task needs more

Pick by the job below. Every model on this board is served through the same key and balance, at the rates on the rate card.

Air's board row → The free tier on the pricing page →

Air has no web search and no reasoning dial. Air's board row mirrors the model it points at and may briefly lag an alias change.

Agentic coding

Many steps, many files, and the model has to keep its own plan straight for an hour. A one-shot index only half predicts that, so run your own repo through the top two.

Pick

GPT 5.6 Sol orClaude Opus 5

On a budget

DeepSeek V4 Pro $0.67/task at top effort
The genuinely hard problems

Research, analysis and anything where being wrong is expensive.

Pick

Claude Fable 5.1 orClaude Opus 5

On a budget

GPT 5.6 Sol $1.99/task at top effort
Anything a person waits on

Chat, voice, autocomplete. The first token has to land before attention does.

Pick

Gemini 3.7 Flash orGPT 5.6 Luna
Documents, images, audio, video

One request carrying a whole corpus of mixed media, reasoned over as a unit.

Pick

Gemini 3.7 Flash orMuse Spark 1.2

On a budget

Gemini 3.1 Pro Preview $0.67/task at top effort
High volume, tight budget

Classification, extraction, tagging — millions of calls where the bill is the constraint.

Pick

GPT 5.6 Luna orGPT 5.4 Nano
Weights you can run yourself

Data that cannot leave your environment, or volume that beats renting.

Pick

GLM 5.3 orKimi K3

On a budget

GLM 5.3 Flash $0.25/task at top effort

Leaderboard

Compare the full catalog.

Sort and filter the table, then expand a row for context, modalities, reasoning settings, and billing notes.

The models MindsHub serves, ranked by quality index, with a coding read, output speed, latency, measured cost per task at top reasoning effort, price per million tokens, context window and weight availability.

OpenAI `gpt-6-astra`	53 Frontier	—	68 tok/sec	—	$3.26	$10 / $50 our rate	1M	Closed
Hard end-to-end work across code, browsers, and professional files Pick it when The job is a long, tool-driven run — a browser to operate, professional files to work through, a complicated task to finish with little guidance — and you want a premium specialist for it rather than a general flagship. Skip it when Drafting, summaries and routine analysis: the GPT 5.6 family covers that at a fraction of the rate card, with Sol a few points behind on the index. Inputs TextImages API alias `gpt-6-astra` `gpt` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowmediumhighxhighmax` We send medium unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Worth knowing The top notch buys nothing on the score here. This row quotes max; one notch down at xhigh Artificial Analysis measures the same 53 for $2.31 a task instead of $3.26. Read the score as a ceiling and the cost as a worst case.
Anthropic `fable-5-1`	53 Frontier	77	70 tok/sec	3.6s first token	$7.63	$10 / $50 our rate	1M	Closed
Anthropic's newest frontier model, safeguard fallback included Pick it when The problem is at the edge of what any model clears, the run is long, and being wrong costs more than the premium rate does. Skip it when Routine work, or anything a cheaper frontier row already clears — Opus 5 sits two points back at $5.86 a task against $7.63, and GPT 5.6 Sol six points back at about a quarter of the cost. Inputs TextImages API alias `fable-5-1` `fable` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowmediumhighxhighmax` We send high unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Worth knowing "With fallback" is Artificial Analysis's label for the API's default configuration: the small share of prompts Anthropic's safeguard classifiers flag — cybersecurity and biology, fewer than 5% of sessions by Anthropic's count — is answered by a fallback Claude model, and the score above includes those answers. The dial matters as much as the model here: at xhigh Artificial Analysis measures the same 53 for $5.98 a task, at high 51 for $3.91 — Opus 5's score for a third less — and at medium 49 for $2.98, a point below Fable 5 at about a third of its cost.
Anthropic `opus-5`	51 Frontier	72.4	62 tok/sec	3.2s first token	$5.86	$5 / $25 our rate	1M	Closed
Frontier reasoning built for long, many-step agentic runs Pick it when The task is hard, runs for many steps, or the output carries risk that someone has to sign off on. Skip it when You're paying by the token for routine work — it keeps chewing long after a cheaper model would have stopped. Inputs TextImages API alias `opus-5` `opus` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowmediumhighxhighmax` We send high unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Worth knowing The top notch of its dial buys one point. This row quotes max; one notch down at xhigh Artificial Analysis measures 50 for $4.88 a task instead of $5.86, and at high 48 for $3.61. Read the score as a ceiling and the cost as a worst case.
Anthropic `fable-5`	50 Frontier	75	70 tok/sec	3.6s first token	$8.75	$10 / $50 our rate	1M	Closed
The previous Fable, for traffic already validated against it Pick it when You have production traffic validated against this version and re-testing it on 5.1 is not yet worth the interruption. Skip it when You are starting fresh — Fable 5.1 scores three points higher on the same list price, and at high it still clears this row's 50 by a point for $3.91 a task against $8.75. Inputs TextImages API alias `fable-5` Names exactly this model version and never moves. Reasoning effort `lowmediumhighxhighmax` We send high unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
Meta `muse-spark-1-3`	48 Frontier	—	344 tok/sec	—	$1.60	$1.25 / $4.25 our rate	1M	Closed
Long-context agentic coding that reads text, images and video Pick it when Long agent and coding runs over mixed inputs, where a fast closed model at a low rate beats owning the weights and Meta is an acceptable processor of your data. Skip it when You need weights to host — the Muse family is closed — or your data cannot go to Meta, and the Contributor discount does not change that: it is paid for with your prompts. Inputs TextImagesVideo API alias `muse-spark-1-3` `muse-spark` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort Single setting — no dial. Nothing to turn down, so the columns above are what you get. That also makes it predictable to budget. Worth knowing Two rate cards. Artificial Analysis costs it at Meta's Standard rate, which is the one this row prints; the Contributor tier is far cheaper in exchange for training on your prompts and replies, and has no published cost per task. Fine for hobby projects, an easy no for anything confidential.
OpenAI `gpt-5-6-sol`	47 Frontier	75.6	70 tok/sec	2.0s first token	$1.99	$5 / $30 our rate	1M	Closed
OpenAI's flagship all-rounder Pick it when You want one provider that is competent at everything, with the deepest tooling ecosystem around it. Skip it when Price is the binding constraint — GPT 5.6 Terra scores five points lower for about 30% less per finished task. Inputs TextImages API alias `gpt-5-6-sol` Names exactly this model version and never moves. Reasoning effort `nonelowmediumhighxhighmax` We send medium unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
Zhipu via Fireworks AI `glm-5-3`	45 Frontier	65	68 tok/sec	1.6s first token	$2.01	$1.40 / $4.40 our rate	1M	Open GLM-5.3 licence (commercial use restricted)
Frontier-class coding and security review at a low rate Pick it when You want strong coding and security review at a fraction of frontier prices, rented or on your own hardware. Skip it when You need image input, or a permissive licence — this one is Zhipu's own, with commercial restrictions. Inputs Text API alias `glm-5-3` `glm` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowhighmax` We send max unless you ask for another, and that is this model's top setting — the one Quality and Cost/task above were measured at. Turning it down is the saving on offer here, not a surprise on the invoice. Licence GLM-5.3 licence (commercial use restricted) Served by Fireworks AI Zhipu makes it; we buy the inference from Fireworks AI, which is the name the rate card shows. Worth knowing Zhipu released the weights at the end of August under its own GLM-5.3 licence, which allows commercial use with restrictions — read the terms before you build on it. A point ahead of Kimi K3 on the index at the same cost per task to within a cent, and it streams faster; Kimi takes images, this does not.
Moonshot AI via Fireworks AI `kimi-k3`	44 Near-frontier	60.2	42 tok/sec	3.4s first token	$2.00	$3 / $15 our rate	1M	Open Kimi K3 licence (commercial use restricted)
Long autonomous coding runs you can self-host Pick it when You want one of the strongest open-weights models with image input as well as text — GLM-5.3 sits a point ahead on the index but reads text only. Skip it when The token rate is the point — GLM-5.3 charges well under half per token for a point more on the index, though the two finish a task for about the same. Inputs TextImages API alias `kimi-k3` `kimi` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort Single setting — no dial. Nothing to turn down, so the columns above are what you get. That also makes it predictable to budget. Licence Kimi K3 licence (commercial use restricted) Served by Fireworks AI Moonshot AI makes it; we buy the inference from Fireworks AI, which is the name the rate card shows. Worth knowing The licence is Moonshot's own, not MIT or Apache. Self-hosting and fine-tuning are fine; reselling access at scale needs a read.
SpaceXAI `grok-4-6`	44 Near-frontier	60	68 tok/sec	1.2s first token	$2.32	$2 / $6 our rate	500K	Closed
Reasoning with access to the live web Pick it when You need information from this morning in the same call as the reasoning. Skip it when Your agents live in a terminal, or your prompts are enormous — the window is half what most of this board offers. Inputs TextImages API alias `grok-4-6` `grok` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowmediumhighxhigh` We send high unless you ask for another. Quality and Cost/task above were measured at xhigh, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Worth knowing Its dial stops paying off at the top. This row quotes xhigh, the highest setting Artificial Analysis measures, and one notch down at high the same model scores the same 44 for a fifth less per task. What the top setting buys is two tokens a second.
Zhipu via Fireworks AI `glm-flash-5-3`	42 Near-frontier	62	110 tok/sec	1.5s first token	$0.25	$0.15 / $0.50 our rate	1M	Open MIT
A high score per dollar, on MIT-licensed weights Pick it when You want a high score per dollar on MIT weights, with the image input the GLM flagship lacks. Skip it when Your prompts are short and interactive — it answers at length, so a finished task costs more than its rate card suggests. Inputs TextImages API alias `glm-flash-5-3` `glm-5-3-flash` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowhighmax` We send max unless you ask for another, and that is this model's top setting — the one Quality and Cost/task above were measured at. Turning it down is the saving on offer here, not a surprise on the invoice. Licence MIT Served by Fireworks AI Zhipu makes it; we buy the inference from Fireworks AI, which is the name the rate card shows. Worth knowing Artificial Analysis measures it as very verbose, so a finished task uses more tokens, and costs more, than its rate card suggests. First natively multimodal model in the GLM-5 line — 320B total parameters, 18B active.
OpenAI `gpt-terra-5-6`	42 Near-frontier	66	118 tok/sec	1.3s first token	$1.40	$2 / $12 our rate	1M	Closed
Balanced mid-tier for everyday work Pick it when You want the sensible default: strong enough for most work, and it keeps most of its discount on the invoice. Skip it when The last few index points decide the outcome. Inputs TextImages API alias `gpt-terra-5-6` `gpt-terra` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `nonelowmediumhighxhighmax` We send medium unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
Google `gemini-flash-3-8`	41 Strong	—	343 tok/sec	—	$1.24	$0.75 / $3.75 our rate	1M	Closed
Fast multimodal agents and long coding work Pick it when Mixed-media inputs — text, images, audio, video — feeding an interactive agent or a long coding session, where the answer has to come back quickly. Skip it when The problem is genuinely hard: its reasoning sits below the leaders, and Google itself says it can spend more tokens and tool calls than 3.7 Flash on difficult work. Budget with the post-introductory price in mind. Inputs TextImagesAudioVideo API alias `gemini-flash-3-8` `gemini-flash` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowmediumhigh` We send medium unless you ask for another. Quality and Cost/task above were measured at high, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
DeepSeek via Fireworks AI `deepseek-v4-1-flash`	40 Strong	—	227 tok/sec	—	$0.27	$0.22 / $0.66 our rate	1M	Open MIT
Low-cost text and image work on MIT-licensed weights Pick it when High-volume text and image work that still wants step-by-step reasoning, on weights you can take with you. Skip it when You need audio or video input, or the last few points of quality. Inputs TextImages API alias `deepseek-v4-1-flash` `deepseek-v4-flash` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `nonelowhighmax` We send high unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Licence MIT Served by Fireworks AI DeepSeek makes it; we buy the inference from Fireworks AI, which is the name the rate card shows.
Meta `muse-spark-1-2`	40 Strong	54	211 tok/sec	0.7s first token	$0.97	$1.25 / $4.25 our rate	1M	Closed
Cheap closed model that powers the free Meta AI Pick it when You want a fast, capable, every-input model and Meta is an acceptable processor of your data. Skip it when You need weights — the Muse family is closed — or your data cannot go to Meta. Inputs TextImagesAudioVideo API alias `muse-spark-1-2` Names exactly this model version and never moves. Reasoning effort Single setting — no dial. Nothing to turn down, so the columns above are what you get. That also makes it predictable to budget. Worth knowing Meta publishes a "contributor" tier at roughly a fifteenth of the standard rate in exchange for training on your prompts and replies. Fine for hobby projects, an easy no for anything confidential.
Alibaba via Fireworks AI `qwen-3-8-a95b`	40 Strong	56	40 tok/sec	4.1s first token	$2.16	$2 / $6 our rate	984K	Open Qwen licence (commercial use restricted)
The open checkpoint that matches the Qwen flagship Pick it when You need the weights, and text-only input plus a restrictive licence are acceptable. Skip it when You are reselling access, or your inputs are not text. Inputs Text API alias `qwen-3-8-a95b` `qwen` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowmediumxhigh` We send xhigh unless you ask for another, and that is this model's top setting — the one Quality and Cost/task above were measured at. Turning it down is the saving on offer here, not a surprise on the invoice. Licence Qwen licence (commercial use restricted) Served by Fireworks AI Alibaba makes it; we buy the inference from Fireworks AI, which is the name the rate card shows. Worth knowing Artificial Analysis costs it at Alibaba's own $2 / $6 — the same rate card as the proprietary Max — but the weights are downloadable, so your actual bill is whatever your host or your hardware charges.
Google `gemini-flash-3-7`	39 Strong	72	335 tok/sec	0.5s first token	$0.93	$0.75 / $3.75 our rate	1M	Closed
Fast multimodal work at a promotional price Pick it when Inputs are large or mixed-media — PDFs, images, audio, video — and you want them read quickly. Skip it when You need frontier-level reasoning, or weights you can host yourself. Inputs TextImagesAudioVideo API alias `gemini-flash-3-7` Names exactly this model version and never moves. Reasoning effort `lowmediumhigh` We send medium unless you ask for another. Quality and Cost/task above were measured at high, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
SpaceXAI `grok-4-5`	39 Strong	56	60 tok/sec	1.4s first token	$1.04	$2 / $6 our rate	500K	Closed
The previous Grok, and now the better-value one Pick it when You want most of 4.6 for much less per task, or you have production traffic validated against this version. Skip it when You need the five index points 4.6 adds. Inputs TextImages API alias `grok-4-5` Names exactly this model version and never moves. Reasoning effort `lowmediumhigh` We send high unless you ask for another, and that is this model's top setting — the one Quality and Cost/task above were measured at. Turning it down is the saving on offer here, not a surprise on the invoice.
Zhipu via Fireworks AI `glm-5-2`	39 Strong	62	62 tok/sec	1.8s first token	—	$1.40 / $4.40 our rate	1M	Open MIT
The previous GLM release, with the same coding profile Pick it when You want the GLM coding profile and can spare the points 5.3 adds. Skip it when 5.3 is about six index points better on the same rate card, and its weights are listed as open too. Inputs Text API alias `glm-5-2` Names exactly this model version and never moves. Reasoning effort `highmax` We send max unless you ask for another, and that is this model's top setting — the one Quality and Cost/task above were measured at. Turning it down is the saving on offer here, not a surprise on the invoice. Licence MIT Served by Fireworks AI Zhipu makes it; we buy the inference from Fireworks AI, which is the name the rate card shows. Worth knowing MIT weights, where 5.3 ships under Zhipu's own licence with commercial restrictions — so if the terms decide it, this is the GLM you can build on without a read. Artificial Analysis scores this version only provisionally on the current index and publishes no cost per task for it, so both published columns here are our own read.
MindsHub `mindshub_air` Free tier	38 Solid	48	120 tok/sec	0.5s first token	$0.18	$0.20 / $1.20 our rate	1M	Closed
Our alias, kept on whichever model balances best Pick it when You want a good default without re-picking a model every time the field moves. Point at the alias and we keep it on the model that balances cost, speed and intelligence best. Skip it when You are benchmarking, pinning a version, or otherwise need the numbers in this row to hold still. Name a model directly instead. Inputs TextImages API alias `mindshub_air` A rotating alias: we move it as better options ship, so the model behind it can change without a code change on your side. Reasoning effort Single setting — no dial. Nothing to turn down, so the columns above are what you get. They are mirrored figures, though — see the caveat below before you budget on them. Worth knowing Every figure on this row is mirrored from whatever model the alias resolves to today, so it can match another row on this board exactly — and it moves when we move the alias, without notice and with no version to pin. Name a model directly if you need the numbers to stay put.
OpenAI `gpt-luna-5-6`	38 Solid	48	120 tok/sec	0.5s first token	$0.18	$0.20 / $1.20 our rate	1M	Closed
Fast and cheap for high-volume work Pick it when Volume is high, tasks are routine, and the bill matters more than the last few points of quality. Skip it when The work is genuinely hard. It scores alongside good open models, not alongside the frontier. Inputs TextImages API alias `gpt-luna-5-6` `gpt-luna` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `nonelowmediumhighxhighmax` We send medium unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
Anthropic `sonnet-5`	38 Solid	70	89 tok/sec	2.4s first token	$5.09	$2 / $10 our rate	1M	Closed
The lighter Claude, for simpler and shorter tasks Pick it when You want Claude's tone and care on routine drafting, summaries and small edits. Skip it when Anything hard or high-volume: Opus 5 scores thirteen points higher, and per finished task Sonnet costs more than its rate card suggests. Inputs TextImages API alias `sonnet-5` `sonnet` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowmediumhighmax` We send high unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Worth knowing Twice as cheap as GPT 5.6 Sol per output token and about two and a half times the cost per finished task — a plain case of a rate card not predicting the bill. Its streaming rate moves with effort too: 89 tok/s at the top setting these figures come from, against 66 at the high setting we send by default.
OpenAI `gpt-codex-5-3`	38 Solid	68	139 tok/sec	1.7s first token	—	$1.75 / $14 our rate	400K	Closed
Coding specialist for repo-scale changes Pick it when The job is software, across many files, and you want a model tuned for that rather than for general reasoning. Skip it when The work is not code. Its general index score sits below its coding score for a reason. Inputs TextImages API alias `gpt-codex-5-3` `gpt-codex` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowmediumhighxhigh` We send medium unless you ask for another. Quality and Cost/task above were measured at xhigh, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Worth knowing Artificial Analysis scores this version only provisionally — on an older component set that omits both of the benchmarks they run on every fully scored row (Terminal-Bench v4.0 and GDPval-AA v2) — and publishes no cost per task for it, as they do for none of their provisional scores. Their figure is not on the same scale as the rest of this column, so both published columns here are ours.
Meta `muse-spark-1-1`	38 Solid	52	150 tok/sec	0.8s first token	—	$1.25 / $4.25 our rate	1M	Closed
Every input type, at an open-model rate Pick it when You want every input type at a low rate and a fast first token. Skip it when 1.2 is about two index points better on the same rate card. Inputs TextImagesAudioVideo API alias `muse-spark-1-1` Names exactly this model version and never moves. Reasoning effort Single setting — no dial. Nothing to turn down, so the columns above are what you get. That also makes it predictable to budget.
DeepSeek via Fireworks AI `deepseek-v4-pro`	36 Solid	58.6	75 tok/sec	2.6s first token	$0.67	$1.32 / $3.96 our rate	1M	Open MIT
Open-weight reasoning and coding at a budget rate Pick it when Volume is high, the budget is tight, and you want permissive weights as an exit route. Skip it when You need multimodal input or the last few points of quality. Inputs Text API alias `deepseek-v4-pro` `deepseek` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `nonelowhighmax` We send high unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Licence MIT Served by Fireworks AI DeepSeek makes it; we buy the inference from Fireworks AI, which is the name the rate card shows.
DeepSeek via Fireworks AI `deepseek-v4-flash-0731`	35 Solid	52	239 tok/sec	0.9s first token	$0.22	$0.22 / $0.66 our rate	1M	Open MIT
The previous DeepSeek Flash, MIT-licensed and self-hostable Pick it when An integration already pinned to the 0731 build that you are not ready to move. Skip it when You want the current Flash — DeepSeek V4.1 scores higher on the same open licence. Inputs Text API alias `deepseek-v4-flash-0731` Names exactly this model version and never moves. Reasoning effort `nonelowhighmax` We send high unless you ask for another. Quality and Cost/task above were measured at max, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Licence MIT Served by Fireworks AI DeepSeek makes it; we buy the inference from Fireworks AI, which is the name the rate card shows.
Google `gemini-flash-3-6`	34 Solid	68.8	216 tok/sec	0.6s first token	$0.93	$0.75 / $3.75 our rate	1M	Closed
Fast, high-volume everyday tasks Pick it when You want multimodal input and a million-token window at a throughput price. Skip it when 3.7 Flash scores five index points better and is stronger at code. Inputs TextImagesAudioVideo API alias `gemini-flash-3-6` Names exactly this model version and never moves. Reasoning effort `minimallowmediumhigh` We send medium unless you ask for another. Quality and Cost/task above were measured at high, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
Google `gemini-flash-3-5`	33 Light	—	210 tok/sec	1.0s first token	$1.56	$1.50 / $9 our rate	1M	Closed
The legacy Google Flash, for workloads already built on it Pick it when You already run on Gemini 3.5 Flash and want the same model behind the same alias while you evaluate a newer Flash. Skip it when You are starting fresh: Google itself files this as its legacy Flash, and the newer Flash generations here score higher at a lower rate card. Inputs TextImagesAudioVideo API alias `gemini-flash-3-5` Names exactly this model version and never moves. Reasoning effort `minimallowmediumhigh` We send medium unless you ask for another. Quality and Cost/task above were measured at high, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower. Worth knowing Figures are at high, the top of its dial. Artificial Analysis scores its medium and minimal runs only provisionally on the current index and publishes no cost per task for them, so high is the one comparable reading and the one this row quotes.
OpenAI `gpt-mini-5-4`	32 Light	50	150 tok/sec	0.7s first token	—	$0.75 / $4.50 our rate	400K	Closed
Small and cheap for routine automation Pick it when Classification, extraction, tagging and other high-frequency jobs with a narrow definition of correct. Skip it when Anything open-ended. Luna is both cheaper and better at this point in the line-up. Inputs TextImages API alias `gpt-mini-5-4` `gpt-mini` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `nonelowmediumhighxhigh` We send none unless you ask for another — the lowest setting it has, and no reasoning tokens at all. Quality and Cost/task above were measured at xhigh, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
Google `gemini-3-1-pro`	30 Light	68.8	120 tok/sec	1.6s first token	$0.67	$2 / $12 our rate	1M	Closed
Huge documents, images, audio and video Pick it when A single request has to carry a whole corpus of mixed media, and it has to be reasoned over rather than skimmed. Skip it when You are paying by the token — the Flash line reads the same inputs at a third of the rate card, and 3.7 Flash outscores this by nine points. Inputs TextImagesAudioVideo API alias `gemini-3-1-pro` `gemini` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `lowmediumhigh` We send high unless you ask for another, and that is this model's top setting — the one Quality and Cost/task above were measured at. Turning it down is the saving on offer here, not a surprise on the invoice.
Google `gemini-flash-3`	26 Light	—	180 tok/sec	0.9s first token	—	$0.50 / $3 our rate	1M	Closed
Preview-era Flash, kept so early integrations keep running Pick it when A workload already built on the Gemini 3 preview that you are not ready to move, and want to keep calling by its own alias. Skip it when Anything new: it is a preview from December 2025, and each Flash release since scores higher here. Inputs TextImagesAudioVideo API alias `gemini-flash-3` Names exactly this model version and never moves. Reasoning effort `minimallowmediumhigh` We send high unless you ask for another, and that is this model's top setting — the one Quality and Cost/task above were measured at. Turning it down is the saving on offer here, not a surprise on the invoice. Worth knowing Artificial Analysis lists it in reasoning and non-reasoning configurations and publishes no cost per task for either — on their table, the mark of a score not yet on the current component set — so both published columns here are our own read, with the Quality figure carrying their provisional reasoning-mode score. Through MindsHub it exposes a minimal-to-high dial.
OpenAI `gpt-nano-5-4`	22 Basic	30	220 tok/sec	0.4s first token	—	$0.20 / $1.25 our rate	400K	Closed
Built for throughput on bulk, simple work Pick it when Millions of tiny calls where latency is the product and the task fits in one sentence. Skip it when Any judgement is required. Inputs Text API alias `gpt-nano-5-4` `gpt-nano` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `nonelowmediumhighxhigh` We send none unless you ask for another — the lowest setting it has, and no reasoning tokens at all. Quality and Cost/task above were measured at xhigh, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
Anthropic `haiku-4-5`	18 Basic	43.4	101 tok/sec	1.1s first token	$0.21	$1 / $5 our rate	200K	Closed
Fastest Claude, for high-volume simple work Pick it when You want Claude behaviour and Claude tooling on work that does not need Claude-sized thinking. Skip it when The job needs reasoning headroom — this is the previous generation, and it shows on hard tasks. Inputs TextImages API alias `haiku-4-5` `haiku` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort Single setting — no dial. Nothing to turn down, so the columns above are what you get. That also makes it predictable to budget.
Google `gemini-flash-lite-3-1`	16 Basic	—	288 tok/sec	—	$0.04	$0.25 / $1.50 our rate	1M	Closed
Lite-tier extraction and classification at scale Pick it when High-volume, simple work — extraction, classification, routing, transcription — where per-call cost and latency matter more than depth. Skip it when Multi-step reasoning or coding — the Flash and Pro lines exist for that. Inputs TextImagesAudioVideo API alias `gemini-flash-lite-3-1` `gemini-flash-lite` The first names exactly this model and never moves. The second is the rotating alias for its line: it points here today and follows the line when we move it. Both reach the same model — see Rotating aliases. Reasoning effort `minimallowmediumhigh` We send minimal unless you ask for another — the lowest setting it has, and no reasoning tokens at all. Quality and Cost/task above were measured at high, this model's top setting, so this row's board position is a ceiling rather than its out-of-the-box behaviour: at our default you pay less, and on hard work you may score lower.
Cerebras `mindshub_blaze`	—	—	—	—	—	$0.35 / $0.75 our rate	—	Closed
Pick it when Not rated on this board. Skip it when Not rated on this board. Inputs Text API alias `mindshub_blaze` A rotating alias: we move it as better options ship, so the model behind it can change without a code change on your side. Reasoning effort Single setting — no dial. Nothing to turn down, so the columns above are what you get. That also makes it predictable to budget.
Fireworks AI Alternate host `gpt-oss-fireworks`	—	—	—	—	—	$0.15 / $0.60 our rate	—	Closed
Pick it when Not rated on this board. Skip it when Not rated on this board. Inputs Text API alias `gpt-oss-fireworks` The same line served through another host; its rotating alias is `mindshub_blaze`. Reasoning effort Single setting — no dial. Nothing to turn down, so the columns above are what you get. That also makes it predictable to budget.

Quality & Cost/task: Artificial Analysis's published figures, on Intelligence Index v4.3 — they measure independently of us — at each model's top reasoning effort, so both are a ceiling. Expand a row for the level we send.
Coding, Speed & Latency: Speed is Artificial Analysis's published median where they publish one; coding and latency are our own read. Close values are ties, not differences.
Price: Per 1M tokens, in / out. Everything else about the bill — prompt-cache rates, long-prompt tiers and search — is on the full rate card.
Missing figures: ≈ is our estimate. — means nobody has published one.
Checked: Quality and cost per task against Artificial Analysis's table on September 8, 2026; list prices and speeds last refreshed from their API on September 14, 2026. Our rates are from a committed snapshot right now.

Rotating aliases

One alias per line, pointed at the model we serve for it.

A rotating alias such as gpt always resolves to the most recent version of that line we serve, so it can move without a code change on your side. Every other model on the board also has a fixed alias that names one exact version and never moves; MindsHub Air is the one rotating alias with no fixed counterpart, by design.

Rotating alias	Currently serves	Fixed alias	Older versions still served
`mindshub_air`	Whichever model balances cost, speed and intelligence best — see its board row.	—	—
`mindshub_blaze`	MindsHub Blaze Also served via Fireworks AI.	—	—
`sonnet`	Claude Sonnet 5	`sonnet-5`	—
`opus`	Claude Opus 5	`opus-5`	—
`fable`	Claude Fable 5.1	`fable-5-1`	Claude Fable 5
`haiku`	Claude Haiku 4.5	`haiku-4-5`	—
`gpt`	GPT-6 Astra	`gpt-6-astra`	GPT 5.6 Sol
`gpt-terra`	GPT 5.6 Terra	`gpt-terra-5-6`	—
`gpt-luna`	GPT 5.6 Luna	`gpt-luna-5-6`	—
`gpt-codex`	GPT 5.3 Codex	`gpt-codex-5-3`	—
`gpt-mini`	GPT 5.4 Mini	`gpt-mini-5-4`	—
`gpt-nano`	GPT 5.4 Nano	`gpt-nano-5-4`	—
`gemini`	Gemini 3.1 Pro Preview	`gemini-3-1-pro`	—
`gemini-flash`	Gemini 3.8 Flash	`gemini-flash-3-8`	Gemini 3.7 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash, Gemini 3 Flash Preview
`gemini-flash-lite`	Gemini 3.1 Flash-Lite	`gemini-flash-lite-3-1`	—
`kimi`	Kimi K3	`kimi-k3`	—
`deepseek`	DeepSeek V4 Pro	`deepseek-v4-pro`	—
`deepseek-v4-flash`	DeepSeek V4.1 Flash	`deepseek-v4-1-flash`	DeepSeek V4 Flash
`qwen`	Qwen3.8-2.4T-A95B	`qwen-3-8-a95b`	—
`glm`	GLM 5.3	`glm-5-3`	GLM 5.2
`glm-5-3-flash`	GLM 5.3 Flash	`glm-flash-5-3`	—
`muse-spark`	Muse Spark 1.3	`muse-spark-1-3`	Muse Spark 1.2, Muse Spark 1.1
`grok`	Grok 4.6	`grok-4-6`	Grok 4.5

01 · Quality

Small score differences are ties.

The top 4 scorers sit within 3 points of each other, which is close enough to be noise. When quality is that level, speed and cost decide.

A benchmark averages many tasks and can hide weaknesses that matter to you. Use the scores to shortlist two or three models, then test them on your own work. The method below shows how.

Each score is for a model at a specific reasoning setting, usually the top one. Treat it as a ceiling: the same model at a lower setting scores lower and costs less per task.

What each model family offers Read the benchmark behind the score Choose for your workload Point a coding agent at any of them

Quality index from Artificial Analysis for the full board.

GPT-6 Astra 53
Claude Fable 5.1 53
Claude Opus 5 51
Claude Fable 5 50
Muse Spark 1.3 48
GPT 5.6 Sol 47
GLM 5.3 45
Kimi K3 44
Grok 4.6 44
GLM 5.3 Flash 42
GPT 5.6 Terra 42
Gemini 3.8 Flash 41
DeepSeek V4.1 Flash 40
Muse Spark 1.2 40
Qwen3.8-2.4T-A95B 40
Gemini 3.7 Flash 39
Grok 4.5 39
GLM 5.2 ≈39
MindsHub Air ≈38
GPT 5.6 Luna 38
Claude Sonnet 5 38
GPT 5.3 Codex ≈38
Muse Spark 1.1 ≈38
DeepSeek V4 Pro 36
DeepSeek V4 Flash 35
Gemini 3.6 Flash 34
Gemini 3.5 Flash 33
GPT 5.4 Mini ≈32
Gemini 3.1 Pro Preview 30
Gemini 3 Flash Preview ≈26
GPT 5.4 Nano ≈22
Claude Haiku 4.5 18
Gemini 3.1 Flash-Lite 16

Bars run from zero and are relative to the leader — a truncated axis would manufacture a gap that isn't there. Scores within 3 points of each other are a tie.

02 · Performance

Measure the wait and the streaming speed.

Latency is the wait for the first token. Throughput is how quickly the rest arrives. The chart shows both because a model can be strong on one and weak on the other.

From the request to a finished answer. Pale is the wait before the first token, solid is the time to stream the rest. A ranking, not a stopwatch: the streaming rate is Artificial Analysis's published median wherever they publish one, the first-token wait is our own read on every row. Answer length changes both the stakes and the order. A 200-token reply lands between 1.1s and 9.1s; at 4,000 tokens the same board runs 12s to 104s, with 21 of 28 rows changing places.

Gemini 3.7 Flash 3.5s
GPT 5.4 Nano 4.9s
DeepSeek V4 Flash 5.1s
Gemini 3.6 Flash 5.2s
Muse Spark 1.2 5.4s
Gemini 3.5 Flash 5.8s
Gemini 3 Flash Preview 6.5s
GPT 5.4 Mini 7.4s
Muse Spark 1.1 7.5s
MindsHub Air 8.8s
GPT 5.6 Luna 8.8s
GPT 5.3 Codex 8.9s
GPT 5.6 Terra 9.8s
Gemini 3.1 Pro Preview 9.9s
GLM 5.3 Flash 10.6s
Claude Haiku 4.5 11.0s
Claude Sonnet 5 13.6s
Grok 4.6 15.9s
DeepSeek V4 Pro 15.9s
GPT 5.6 Sol 16.3s
GLM 5.3 16.3s
Claude Fable 5.1 17.9s
Claude Fable 5 17.9s
GLM 5.2 17.9s
Grok 4.5 18.1s
Claude Opus 5 19.3s
Kimi K3 27.2s
Qwen3.8-2.4T-A95B 29.1s

Wait for the first token Streaming the answer

An 8-fold spread in completion time

Gemini 3.7 Flash finishes that answer in about 3 seconds and Qwen3.8-2.4T-A95B takes about 29. Same answer length, same request. Nobody waits 29 seconds for a chat reply, and both of those models are on this board.

Short answers make latency more important

On a 200-token reply the wait is 23% to 56% of the wall clock, so latency is worth weighting. By a few thousand tokens it is a rounding error and throughput decides alone. The control above the chart switches between the two views.

Measure with your own traffic

Region, prompt length, cache hit rate and reasoning effort all move these figures. Use them to rank candidates, then measure the two you shortlist against your own traffic.

03 · Cost

Price per token is not cost per task.

Token prices do not show how many tokens a model will use to finish the job. Reasoning models can vary widely on that second number.

You pay for reasoning tokens as well as the visible answer. A model with a lower token rate can still cost more per completed task.

It is somebody else's task, at each model's top effort, priced at the provider's own rate — so treat it as a way to rank candidates, not a forecast of your bill.

Why cheap models often aren't Two levers that cut the bill The full rate card

How much of the discount survives the invoice. Each model against GPT 5.6 Sol: what its published output rate promises, then what its measured cost per task delivers — both at top effort. These 5 keep the least.

Claude Sonnet 5 2.0× cheaper on paper 2.6× dearer per task
GLM 5.3 Flash 40.0× cheaper on paper 8.0× cheaper per task
GLM 5.3 4.5× cheaper on paper 1.0× dearer per task
Grok 4.6 3.3× cheaper on paper 1.2× dearer per task
Muse Spark 1.3 4.7× cheaper on paper 1.2× cheaper per task

Both figures come from the same rate card — each provider's own published price, which is what these tasks were costed at, and not always the rate in the Price column above. And the lesson is not that cheap models are a lie: Gemini 3.1 Flash-Lite beats its own rate card — 13.3× cheaper on paper, 49.8× cheaper per task. It is that you cannot rank them by reading price pages.

04 · Method

Choose a model in an afternoon.

Use the board for a shortlist, then test it against real examples.

Shortlist two or three from the board

Use the task router above, or filter the board to your constraint — open weights, a modality, a latency ceiling. Stop at three.
Collect ten real tasks

With the messy context attached. Not clean samples: the actual ones you ran last week, including the awkward ones.
Run them at the effort you would really use

Not maximum. The reasoning dial moves cost further than switching models does, and every published cost-per-task figure is measured at the top setting.
Read the token counts off the response

Every provider returns a usage object. Multiply input, cached input and output separately — those three rates can differ by an order of magnitude.
Divide by the results you would actually ship

If eight of ten came back usable, your real cost is the total divided by eight. That division is the one no leaderboard can do for you.

One API key reaches every model here, so a bake-off is a string change rather than three integrations. There's a runnable version, a cost measurement guide, a guide to running evals, and a free working session if you'd rather do it with an engineer.

FAQ

Questions about the rankings.

Which model should I use?

Use the task recommendations on the board. Current picks include GPT 5.6 Sol or Claude Opus 5 for agentic coding, Claude Opus 5 for hard reasoning, Gemini Flash or GPT 5.6 Luna for interactive work, and GLM-5.3 or Kimi K3 when you need open weights. Start with MindsHub Air if you do not yet have a strong preference.

Which models does this page cover?

Every model currently in the MindsHub catalog, and only those models, one row per model. Each row includes the alias that calls it: the fixed alias for that exact version, or, for MindsHub Air, the rotating alias itself.

What is a rotating alias?

A rotating alias such as gpt or sonnet always resolves to the most recent version of that line MindsHub serves, so it can change without a code change on your side. Every other model also has a fixed alias that names one exact version and never moves; MindsHub Air is the one rotating alias with no fixed counterpart, by design. The rotating aliases table on the board maps each rotating alias to the model it serves today and to that fixed alias.

Why rank on cost per task instead of price per token?

Models use different numbers of tokens and reasoning steps to finish the same task. Cost per task captures that difference. Use it to compare candidates, not as a forecast of your own bill.

Where do these benchmark numbers come from?

Quality and task cost come from Artificial Analysis at each model's highest reasoning setting. A “≈” marks a MindsHub estimate. Coding scores and first-token latency are MindsHub assessments. Throughput uses Artificial Analysis medians where available, and prices come from the live MindsHub catalog.

At what reasoning effort were these numbers measured?

Artificial Analysis figures use each model's highest setting. MindsHub often sends a lower default, shown when you expand a row, so actual cost may be lower. MindsHub does not publish its own per-setting scores.

Does this page replace the pricing page?

No. The board shows input and output token prices for comparison. Use the pricing page for cache rates, long-prompt tiers, web search, failover order, and billing details.

Models ranked by quality, speed, and cost.

What are you building?

Start free with the MindsHub Air model.

Agentic coding

The genuinely hard problems

Anything a person waits on

Documents, images, audio, video

High volume, tight budget

Weights you can run yourself

Compare the full catalog.

One alias per line, pointed at the model we serve for it.

Small score differences are ties.

Measure the wait and the streaming speed.

An 8-fold spread in completion time

Short answers make latency more important

Measure with your own traffic

Price per token is not cost per task.

Choose a model in an afternoon.

Shortlist two or three from the board

Collect ten real tasks

Run them at the effort you would really use

Read the token counts off the response

Divide by the results you would actually ship

Questions about the rankings.

Test the shortlist with one key.