MiniMax M3 (batch)
Sold by 7 sellers, timed at 70.7 tok/s, scored 18.9 on the Intelligence Index, on 19 Design Arena boards, and with results on 2 openly licensed boards.
retrieved sha256 · 0bb53a77713f
What sellers charge
7 sellers · 2.46x from cheapest to dearestOne tick per seller, on a log scale of input price per million from $0.224 at DeepInfra to $0.55 at Featherless. The $0.224 listed is the cheapest on offer; 6 sellers are dearer. A faint tick declares no number format. Every seller is named on the endpoint ledger below.
Measured speed
14,609 requests · 8 endpoints · 30-minute windowOpenRouter's own timing of its routed traffic, weighted by requests, over one 30-minute window read . Every route is on the endpoint ledger below.
Every endpoint
7 priced · 8 timed · cheapest first- DeepInfradeepinfra/fp8/flex · listed$0.224$0.88$0.0448fp8131knot matchednonenonenot matched100%
- CoreWeavecoreweave/fp4$0.23$0.96$0.05fp4236k98 tok/s54457ms6,104 req93.67%
- DeepInfradeepinfra/fp8$0.28$1.10$0.056fp8131knot matchednonenonenot matched99.69%
- Parasailparasail/fp8$0.30$1.20$0.06fp8524knot matchednonenonenot matched99.64%
- Venicevenice/fp8$0.30$1.20$0.06fp8131k108 tok/s181.25s834 req94.05%
- Parasailparasail/fast$0.45$1.80$0.09fp8524knot matchednonenonenot matched99.64%
- Featherlessfeatherless/fp8$0.55$2.20$0.06fp833k82 tok/s49.8944ms109 req95.12%
- DeepInfranot matchednot matchednot matchednot matchedno offer43 tok/s231.35s9 reqno offer
- DeepInfranot matchednot matchednot matchednot matchedno offer55 tok/s181.13s201 reqno offer
- Minimaxno priceno priceno priceno offerno offer22 tok/s133.04s4,945 reqno offer
- Parasailnot matchednot matchednot matchednot matchedno offer120 tok/s44639ms695 reqno offer
- Parasailnot matchednot matchednot matchednot matchedno offer77 tok/s41598ms1,712 reqno offer
Each track runs from a route's p90, the rate nine in ten of its requests met or beat, up to its median. Weighted by requests, the figure above reports where the traffic went rather than what one route does: 3 of the 8 timed routes are slower than it, and the middle route runs at 79.5 tok/s. The widest median-to-p90 gap is Venice's, 6x.
DeepInfra publishes 2 offers and had 2 routes timed; Parasail publishes 2 offers and had 2 routes timed. The two documents share no route id, so those offers and routes are printed apart rather than paired.
Uptime is OpenRouter's own figure for each seller over the trailing day, read once a day, and the column prints the latest reading, taken . 12 of the 16 sellers with a reading fell under 99% on at least one of the 23 days held, 23 Oct 2026 to 21 Nov 2026: Featherless at 10.34% on 19 Nov 2026, DeepInfra (deepinfra/fp8/flex) at 20.00% on 11 Nov 2026, Mara at 42.85% on 26 Oct 2026, Venice at 64.95% on 24 Oct 2026, and 8 more. Source: OpenRouter's public endpoints API, one reading a day for each seller: 1,211 sellers carried one on the latest of the 23 days held.
Source: OpenRouter's public endpoints API. 1257 sellers across 359 models when the mirror ran, 186 of them served by more than one. List prices, not what anybody paid; a discount is the seller's own. Read . Speed is OpenRouter's own timing over one 30-minute window, read .
What each seller accepts
19 parameters · 7 sellers- frequency_penaltyevery seller
- include_reasoningevery seller
- max_tokensevery seller
- presence_penaltyevery seller
- reasoningevery seller
- stopevery seller
- temperatureevery seller
- top_kevery seller
- top_pevery seller
- repetition_penalty6 of 7not Venice
- seed6 of 7not Venice
- tool_choice6 of 7not Featherless
- tools6 of 7not Featherless
- logit_bias4 of 7not CoreWeave, Featherless, and Venice
- response_format4 of 7not Featherless, Parasail (parasail/fp8), and Venice
- logprobs3 of 7only CoreWeave, Parasail (parasail/fast), and Parasail (parasail/fp8)
- min_p3 of 7only DeepInfra (deepinfra/fp8), DeepInfra (deepinfra/fp8/flex), and Featherless
- top_logprobs3 of 7only CoreWeave, Parasail (parasail/fast), and Parasail (parasail/fp8)
- structured_outputs2 of 7only CoreWeave and Parasail (parasail/fast)
9 of these 19 are accepted by every one of the 7 sellers that publish a list; the other 10 are declared by some sellers and not others, so what a request sending one of them gets depends on which seller serves it. Each list is the seller's own declaration in OpenRouter's endpoints API, read , and nothing here tested one.
Price position
342 priced modelsBlended $0.388 per million tokens, 3 in to 1 out: cheaper than 233 of the 342 priced models in the catalog. Each bar is one stretch of the catalog on a logarithmic price scale; the mark is this listing.
Measured quality
145 models with an index · 132 on a design arena boardEach mark is one board this model stands on, placed at the share of that board it scores above, and led to its number in the list under the rule. The columns and the ledger below carry the figures.
- Agentic game development · Agents1112 Elowin rate 44.7% · source rank 18the board, 25 ranked
- Android native · Agents1155 Elowin rate 44.9% · source rank 26the board, 41 ranked
- blender3d · Agents1066 Elowin rate 44.0% · source rank 14the board, 25 ranked
- codetovideo · Agents1221 Elowin rate 55.8% · source rank 9the board, 26 ranked
- Full stack · Agents1175 Elowin rate 47.5% · source rank 20the board, 44 ranked
- HTML slides · Agents1077 Elowin rate 46.9% · source rank 14the board, 25 ranked
- Mobile apps · Agents1169 Elowin rate 46.8% · source rank 30the board, 49 ranked
- pdf · Agents1267 Elowin rate 60.8% · source rank 7the board, 36 ranked
- python-pptx slides · Agents1168 Elowin rate 47.2% · source rank 16the board, 29 ranked
- spreadsheet · Agents1167 Elowin rate 38.1% · source rank 15the board, 32 ranked
- Web apps · Agents1190 Elowin rate 47.7% · source rank 23the board, 47 ranked
- 3D scenes · Models1189 Elowin rate 49.9% · source rank 55the board, 113 ranked
- ASCII art · Models1139 Elowin rate 45.4% · source rank 45the board, 67 ranked
- Code · Models1234 Elowin rate 50.4% · source rank 48the board, 120 ranked
- Data visualisation · Models1229 Elowin rate 51.0% · source rank 45the board, 110 ranked
- Game development · Models1199 Elowin rate 45.1% · source rank 57the board, 120 ranked
- SVG · Models1146 Elowin rate 43.8% · source rank 50the board, 83 ranked
- UI components · Models1233 Elowin rate 50.2% · source rank 48the board, 116 ranked
- Websites · Models1247 Elowin rate 51.5% · source rank 46the board, 126 ranked
Position 81 of the 258 models Epoch has placed, which is not the population of either board above, on a scale anchored at Claude 3.5 Sonnet 130 and GPT-5 150. Epoch states that absolute values are meaningless by themselves and that comparisons between models are what the index is for, that values are not linearly related to benchmark accuracy, and that a highly specialised model may score low while being capable in its own domain.
Benchmark results
2 of 6 boards · 31 results| Task | All runs | Effort | Score |
|---|---|---|---|
| Chess Puzzles | 167 | none | 0.08 |
| unstated | 0.17 | ||
| GPQA diamond | 342 | none | 0.81 |
| unstated | 0.91 | ||
| Mystery Game Puzzles | 149 | none | 0.08 |
| unstated | 0.07 | ||
| OTIS Mock AIME 2024-2025 | 319 | none | 0.27 |
| unstated | 0.71 |
Mean accuracy, 0 to 1, on the task the row names, as Epoch AI runs it and averaged over repeated runs. All eighteen of their tasks, including the held-out FrontierMath tiers, named as they name them. The model id carries the effort setting the run used. Epoch AI Benchmarking Hub names these runs MiniMax-M3_none and MiniMax-M3, and the effort beside each score says which run it is, unstated where the publisher wrote none. All runs is how many runs the publisher scored on the task, every model's. Read from epoch.ai ↗, which dates its own data . Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
| Task | Score |
|---|---|
| AMPS_Hard | 75 |
| code_completion | 67.39 |
| code_generation | 69.01 |
| connections | 96.33 |
| consecutive_events | 84.61 |
| integrals_with_game | 56 |
| javascript | 63.64 |
| logic_with_navigation | 66 |
| math_comp | 91.18 |
| olympiad | 85.61 |
| paraphrase | 55.93 |
| plot_unscrambling | 50.17 |
| python | 35 |
| simplify | 56.62 |
| spatial | 94 |
| story_generation | 63.55 |
| summarize | 53.93 |
| tablejoin | 45.85 |
| tablereformat | 98.04 |
| theory_of_mind | 76.92 |
| typescript | 23.33 |
| typos | 84 |
| zebra_puzzle | 61 |
Percent correct on the task the row names, higher is better. All 23 published columns; there is no average to carry, because the site computes its headline in the browser from a weighting this does not reimplement. LiveBench names this run minimax-m3, with no effort setting written on it. Read from livebench.ai ↗, which dates its own data . LiveBench, Apache-2.0. Results produced by the LiveBench maintainers on questions refreshed monthly.
2 of the 6 openly licensed boards Vitna mirrors carry a result for MiniMax M3 (batch), each printed as its publisher published it and none added to another. MTEB, HELM, Vectara hallucination leaderboard and Aider polyglot have not published one under a name that matches its listing, and names are matched to the catalog exactly and never guessed. A task that is a link opens its column on /capability, with every run scored on it; an effort that is a link opens its run's row there, where the run is ranked on Vitna's scale. Every board, with its licence and what its publisher measures, is on /benchmarks.
Source: OpenRouter (openrouter.ai/models), as of 17 Nov 16:29 UTC. sha256 0bb53a77713f over 853,270 bytes.
Catalog: OpenRouter's public model list.
Intelligence, coding and agentic indexes: Artificial Analysis, as published by OpenRouter.
Elo, win rate and rank: Design Arena, as published by OpenRouter.
Source: each model's own Hugging Face repository (huggingface.co), as of 20 Nov 2026. 162 repositories read, 4 did not answer.
Licence, parameter count, model class and gating: each model's own Hugging Face repository, the one OpenRouter names for the listing.
A licence is the tag the uploader put on the repository, read from nothing else and verified by nobody.
A parameter count is the total the repository's safetensors files declare; for a mixture of experts that is every expert, not the share one token runs through.
Source: the Epoch AI Capabilities Index (epoch.ai/eci), as of 20 Nov 2026. 258 models placed over 2917 benchmark results.
Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
Vitna ran none of these evaluations and adjusted none of these figures.
