MiniMax-M3
MiniMaxReleased Jun 1, 2026Open weights · MiniMax Community Licenseminimax-m3
- Intelligence (Epoch)
- 147 · 49th of 111
- Reasoning
- yes
- Knowledge cutoff
- January 2026
- Context
- 1M tokens
- Max output
- 512k tokens
- Input
- text, image, video
- API features
- tools, structured outputs, JSON mode, seed, logprobs
- MiniMax's API
- yes, as MiniMax-M3
Read Oct 23, 06:00 UTC from the hosts' published offers and OpenRouter, spec sheet from MiniMax · cutoff: aiknowledgecutoff.com
The cheapest host, CoreWeave, serves fp4 weights (maker: bf16) and serves a 256K context (maker: 1M). Weights and context per host
Intelligence for the price
5 models score higher for the same price or less; it ranks 6 of the 38 models at its price or below.
Epoch Capabilities Index against the lowest input + output price per M, log scale · prices Oct 23. Score: Epoch AI, CC BY 4.0, 147 with an interval of 142 to 150.
Against its rivals
Its previous version, the best of its price range and the top score
- MiniMax-M3this model$1.19147
- Claude Opus 5.5top score$24.00169
- Claude Haiku 5.5same price range$0.60159
- GPT-6 Lunasame price range$0.60157
- MiniMax M2.7previous version$1.05146
Same level for 7.4× less: Qwen3.7 Flash at Alibaba, $0.16 per M input + output · their Epoch intervals overlap (144 against 147).
Bars: Epoch Capabilities Index (Epoch AI, CC BY 4.0), drawn from 100, below the lowest score in the catalogue · price: lowest input + output per M.
Speed by provider
Output speed of MiniMax-M3 at each host, fastest first
- CoreWeave265 tok/s0.26 s to first token
- AtlasCloud135 tok/s1.40 s to first token
- Parasail104 tok/s0.61 s to first token
- Together AI80 tok/s0.52 s to first token
- Minimax76 tok/s1.20 s to first token
- SambaNova55 tok/s3.22 s to first token
- GMICloud50 tok/s1.97 s to first token
- StreamLake50 tok/s2.32 s to first token
- Venice40 tok/s0.76 s to first token
- DeepInfra28 tok/s1.59 s to first token
Speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · a snapshot of real traffic before the daily read, not a daily average · bars from zero.
Run it yourself
open weights, MiniMax Community License · too large for the 6 workstations in our guideSelf-host guide ›- Parameters
- 427B
- 23B active per token (MoE)
- Memory needed
- 256 GB
- in INT4 · 512 GB in INT8
- Smallest cloud setup
- 4× H100
- 320 GB · $8.60/h, about $6,278 a month
- Worth it above
- 5B tokens
- a month, against the cheapest API at $1.19 per M
| Quantization | Memory | Smallest cloud GPU | Quality |
|---|---|---|---|
| INT4 (Q4) | 256 GB | 4× H100 · $8.60/h | smallest footprint, minor quality loss |
| INT8 (Q8) | 512 GB | 8× H100 · $17.20/h | near-lossless |
| FP16 (full) | 1025 GB | multi-node, over 640 GB | reference quality |
At home: 256 GB in INT4 is beyond any desktop or Mac in our machines table, so it means a multi-GPU server.
Memory ≈ parameters × bytes per parameter × 1.2 · cloud price: on-demand rate, running 24/7 · builds: GGUF, AWQ, MLX, GPTQ, Ollama · break-even calculator
| Provider | Input | Output | Cache ↑ | tok/s | First token | Uptime 24h | Checked |
|---|---|---|---|---|---|---|---|
| GMICloudfp8 (bf16) · 1M ctx · 512K out−20% | $0.24 | $0.96 | $0.048 | 50 | 1.97 s | 99.9% | Oct 23 |
| CoreWeavefp4 (bf16) · 256K ctx (1M) | $0.23 | $0.96 | $0.05 | 265 | 0.26 s | 90.0% | Oct 23 |
| DeepInfrafp8 (bf16) · 512K ctx (1M) · 128K out | $0.28 | $1.10 | $0.056 | 28 | 1.59 s | 100% | Oct 23 |
| Venicefp8 (bf16) · 512K ctx (1M) · 64K out | $0.30 | $1.20 | $0.06 | 40 | 0.76 s | 98.4% | Oct 23 |
| Parasailfp8 (bf16) · 1M ctx · 512K out | $0.30 | $1.20 | $0.06 | 104 | 0.61 s | 100% | Oct 23 |
| AtlasCloudfp8 (bf16) · 524K ctx (1M) | $0.30 | $1.20 | $0.06 | 135 | 1.40 s | 98.8% | Oct 23 |
| Novita AIfp8 (bf16) · 1M ctx · 128K out | $0.30 | $1.20 | $0.06 | — | — | — | Oct 23 |
| Minimaxfp8 (bf16) · 512K ctx (1M) | $0.30 | $1.20 | $0.06 | 76 | 1.20 s | 99.9% | Oct 23 |
| Together AIfp4 (bf16) · 512K ctx (1M) | $0.30 | $1.20 | $0.06 | 80 | 0.52 s | 99.7% | Oct 23 |
| Fireworks AI500K ctx (1M) | $0.30 | $1.20 | $0.06 | — | — | — | Sep 8 |
| StreamLakefp8 (bf16) · 1M ctx · 500K out−50% | $0.30 | $1.20 | $0.06 | 50 | 2.32 s | 98.4% | Oct 23 |
| SambaNovaprecision not stated · 1M ctx | $0.60 | $2.40 | $0.06 | 55 | 3.22 s | 98.6% | Oct 22 |
10 of 12 providers · prices per million tokens, read Oct 23, 06:00 UTC · speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · uptime: OpenRouter, last 24 h · −N%: a discount off the host's list price · under each host: weights precision, context and max output as served, the maker's value in brackets when the host's differs
| Provider | Weights | Context | Max output | Tools | Structured outputs | Reasoning |
|---|---|---|---|---|---|---|
| CoreWeave | fp4 | 256K | not checked yet | yes | yes | yes |
| GMICloud | fp8 | 1M | 512K | yes | no | yes |
| DeepInfra | fp8 | 512K | 128K | yes | no | yes |
| Venice | fp8 | 512K | 64K | yes | no | yes |
| Parasail | fp8 | 1M | 512K | yes | no | yes |
| AtlasCloud | fp8 | 524K | not read: host terms | yes | no | yes |
| Novita AI | fp8 | 1M | 128K | yes | no | yes |
| Minimax | fp8 | 512K | not checked yet | yes | no | yes |
| Together AI | fp4 | 512K | not stated by the host | yes | yes | yes |
| Fireworks AI | not checked yet | 500K | not checked yet | — | — | — |
| StreamLake | fp8 | 1M | 500K | yes | no | yes |
| SambaNova | not stated by the host | 1M | not stated by the host | yes | no | yes |
Weights, context and max output: read at each host's own catalogue or docs where its terms allow, otherwise as the host declares to OpenRouter; hover a cell for its source and date · tools, structured outputs, reasoning: as declared to OpenRouter · not verified by us
MiniMax-M3 spec sheet, field by field, with sources
MiniMax-M3 spec sheetmaker-declared
What MiniMax declares about the model itself — read from its own weights repository and documentation, not from any host. What each endpoint actually serves is in the pricing above: the two can differ, and both are true.
- Context window
- 1,048,576 tokens
- Parameters
- 427,040,140,160≈ 427.0B
- Active experts
- 4 of 128 active per token
- Weight precision
- bfloat16
- License
- MiniMax Community License — open weights, conditionalcurated · read 2026-09-07Scope: the licence on the published weights, and nothing else. It is not the maker's acceptable-use policy, which is a separate document, and it is not the contract you sign with whichever host you call — hosts set their own terms, which this index does not read.
- Knowledge cutoff
- not published by the maker≈ January 2026 third-party estimate — not the maker's figurecurated
- Modalities
- text · image · video
- Max output (official)
- to verify
- Release date
- to verify≈ June 1, 2026 third-party catalogue — not the maker's figure
- Maker lifecycle status
- to verify
Also declared in the weights
- Architecture class
- MiniMaxM3SparseForConditionalGeneration
- Layers
- 60
- Vocabulary
- 200,064 tokens
- Weights first published
- June 2, 2026
- Weights last updated
- July 23, 2026
- Paper
- arXiv:2606.13392
MiniMaxAI/MiniMax-M3 · config.json @ f0e1c1e · read 2026-10-08 · 6 of 10 fields published by the maker
| Date | Provider | Input | Output | Change |
|---|---|---|---|---|
| Oct 22 | DeepInfra | $0.22 → $0.28 | $0.88 → $1.10 | +25% |
| Oct 16 | DeepInfra | $0.28 → $0.22 | $1.10 → $0.88 | −20% |
| Sep 10 | GMICloud | $0.60 → $0.24 | $2.40 → $0.96 | −60% |
Every published change of input or output price, newest first · change on input + output · full series: pricing-history.csv.
| Provider | Chatbot 1,000 in / 500 out | RAG / long context 8,000 in / 500 out | Batch summarize 4,000 in / 1,000 out |
|---|---|---|---|
| CoreWeave | $0.71 | $2.32 | $1.88 |
| GMICloud | $0.72 | $2.40 | $1.92 |
| DeepInfra | $0.83 | $2.79 | $2.22 |
| Venice | $0.90 | $3.00 | $2.40 |
| Parasail | $0.90 | $3.00 | $2.40 |
| AtlasCloud | $0.90 | $3.00 | $2.40 |
| Novita AI | $0.90 | $3.00 | $2.40 |
| Minimax | $0.90 | $3.00 | $2.40 |
| Together AI | $0.90 | $3.00 | $2.40 |
| Fireworks AI | $0.90 | $3.00 | $2.40 |
| StreamLake | $0.90 | $3.00 | $2.40 |
| SambaNova | $1.80 | $6.00 | $4.80 |
What each host bills for 1,000 requests of a representative job
Your own workload
Other MiniMax models · all its models, hosts and history ›
- MiniMax M2.59 providers · from $1.22
- MiniMax M2.77 providers · from $1.05
- MiniMax M3.1 Flash4 providers · from $0.60
- MiniMax M23 providers · from $1.27
- MiniMax M12 providers · from $2.60
- MiniMax M2.12 providers · from $1.50