MiniMax M3
TL;DR
Open-sourceFrontier open-source coding, agents, and native multimodality.
Available on
Open-source models are available on every plan, including Go ($1/mo).
Switch with
Pick MiniMax M3 from the selector.
Intelligence Index
MiniMax M3 vs the MiniMax lineup
MiniMax M3
MiniMax M2.7
MiniMax M2.5
Speed
~113
tokens / sec
Input
$0.30
per M tokens
Output
$1.20
per M tokens
MiniMax M3 deal: 50% off
50% off — permanent. See the deal for details.
| Per 1M tokens | Was | Now |
|---|---|---|
| Standard (≤ 512K) · Input | $0.60 | $0.30 |
| Standard (≤ 512K) · Output | $2.40 | $1.20 |
| Standard (≤ 512K) · Cache read | $0.12 | $0.06 |
| Long context (> 512K) · Input | $0.60 | $0.30 |
| Long context (> 512K) · Output | $2.40 | $1.20 |
| Long context (> 512K) · Cache read | $0.12 | $0.06 |
MiniMax M3 in Command Code
MiniMax M3 is MiniMax's open-source model — frontier coding, agents & native multimodality. It runs in Command Code with a 1M-token context window, switchable any time with /model.
It scores 44.4 on the Intelligence Index — #16 of the 37 scored models in the Command Code lineup.
MiniMax M3 specs at a glance
What MiniMax M3 accepts, how much it can hold in context, and what you need to run it.
| Spec | MiniMax M3 |
|---|---|
| Context window | 1M tokens |
| Input modalities | text + images |
| Reasoning | Yes |
| Release date | 2026-06-01 |
| Minimum plan | Go |
MiniMax M3 vs the Command Code lineup
MiniMax M3 alongside its nearest real alternatives in the lineup — every price straight from the billing tables.
| Model | Intelligence | Coding | Speed | Input $/M | Output $/M | Blended $/M | Context |
|---|---|---|---|---|---|---|---|
| Claude Fable 5 | 59.9 | 76.5 | ~62 tok/s | $10.00 | $50.00 | $20 | 1M |
| Qwen 3.7 Max | 46 | 66 | ~196 tok/s | $1.25 | $3.75 | $1.875 | 1M |
| MiniMax M3 (this page) | 44.4 | 58.6 | ~113 tok/s | $0.30 | $1.20 | $0.525 | 1M |
| DeepSeek V4 Pro | 44.3 | 59.4 | ~62 tok/s | $0.435 | $0.87 | $0.5438 | 1M |
| GPT-5.3 Codex | 44.3 | — | ~106 tok/s | $2.00 | $8.00 | $3.50 | 400K |
| Kimi K2.6 | 44.2 | 61.8 | ~43 tok/s | $0.95 | $4.00 | $1.7125 | 256K |
| Laguna S 2.1 | — | — | — | $0.00 | $0.00 | free | 256K |
What MiniMax M3 is best for
MiniMax M3 earns its slot in the lineup for getting the most out of your credits, long-context work across big codebases.
- Getting the most out of your credits — pick MiniMax M3 from /model and it applies to the session.
- Long-context work across big codebases — pick MiniMax M3 from /model and it applies to the session.
When to switch away from MiniMax M3
No single model wins every task. These are MiniMax M3's computed nearest alternatives — one step up, one step down in cost, one for speed, one from the same family — each switchable mid-session with /model.
Switch to Qwen 3.7 Max
Qwen 3.7 Max scores 46 on the Intelligence Index to MiniMax M3's 44.4 — the nearest genuine step up — at $1.875 blended per million tokens versus $0.525. Reach for it when a task keeps hitting MiniMax M3's ceiling.
Switch to DeepSeek V4 Flash
DeepSeek V4 Flash runs about 67% cheaper blended ($0.175 versus $0.525 per million tokens) while scoring 40.3 on the Intelligence Index. Switch down for high-volume work where MiniMax M3's edge isn't earning its rate.
Switch to Gemini 3.1 Flash Lite
Gemini 3.1 Flash Lite streams ~300 tokens/sec to MiniMax M3's ~113 at a comparable blended cost ($0.5625 per million tokens). Use it when iteration speed matters more than squeezing out the last point of quality.
Switch to MiniMax M2.7
MiniMax M2.7 is the nearest MiniMax sibling — same house style, different price point ($0.525 versus $0.525 blended per million tokens). The natural swap when you want to stay in the family.
What you pay for MiniMax M3
MiniMax M3 is billed per token at the rates below — the same billing tables the Usage page charges against, so this page cannot quote a different price than you pay.
Blended cost (3:1 input:output, the shape of a typical coding session) works out to $0.525 per million tokens.
The rate changes with context size, so each band gets its own row — the flat rate alone would misstate what long-context requests cost.
| Per 1M tokens | Input | Output | Cache read |
|---|---|---|---|
| Context ≤ 512K | $0.30 | $1.20 | $0.06 |
| Context > 512K | $0.30 | $1.20 | $0.06 |
In Command Code: caching and taste-1
Open-source models are routed across multiple upstream providers for high availability. The price you see is the mean per-provider rate; the Usage page reflects what was actually charged.
Where supported by the upstream, prompt caching is on by default — cache reads are billed at $0.06 per million tokens versus $0.30 for fresh input.
taste-1 sits between the model and the agent loop, rewriting and reranking candidate edits to match your codebase conventions.
Plan availability
MiniMax M3 is an open-source model, available on every plan including Go.
Command Code is a subscription with model usage at API rates. Each plan ships with monthly LLM credits; credits roll over and never expire, and auto top-up keeps you running if you go over.
| Plan | Price/mo | LLM credits | Models |
|---|---|---|---|
| Go | $1 | $10 | Open-source only |
| Pro | $15 | $30 | Open-source + premium |
| Provider | $15 | Pay as you go | Open-source + premium |
| Max 10× | $100 | $150 | Open-source + premium |
| Max 20× | $200 | $300 | Open-source + premium |
| Teams Pro | $40 / seat | $40 / seat | Open-source + premium |
| Enterprise | Custom | Custom | Custom pool, SSO, audit logs |
Switching models with /model
In an interactive Command Code session, run /model to open the model selector. Pick MiniMax M3 and it applies to this session and to future sessions until you change it again. Premium models require Pro or higher; open-source models are available on every plan, including Go.
cmd # start an interactive session
/model # open the selector and pick MiniMax M3All Command Code models, ranked by quality and speed
Quality is the Intelligence Index — an aggregate score across reasoning, math, coding, and knowledge evaluations. Speed is measured output tokens per second. Models without a published score are noted. This table is regenerated from the model registry, so it is always current.
| Model | Tier | Intelligence Index | Output speed |
|---|---|---|---|
| Claude Fable 5 | Premium | 59.9 | ~62 tok/s |
| GPT-5.6 Sol | Premium | 58.9 | ~77 tok/s |
| Claude Opus 4.8 | Premium | 55.7 | ~55 tok/s |
| GPT-5.6 Terra | Premium | 55 | ~155 tok/s |
| GPT-5.5 | Premium | 54.8 | ~81 tok/s |
| Grok 4.5 | Open-source | 53.8 | ~114 tok/s |
| Claude Opus 4.7 | Premium | 53.5 | ~52 tok/s |
| Claude Sonnet 5 | Premium | 53.4 | ~79 tok/s |
| GPT-5.4 | Premium | 51.4 | ~164 tok/s |
| GPT-5.6 Luna | Premium | 51.2 | ~234 tok/s |
| GLM-5.2 | Open-source | 51.1 | ~208 tok/s |
| Muse Spark 1.1 | Premium | 50.6 | ~130 tok/s |
| Gemini 3.5 Flash | Premium | 50.2 | ~236 tok/s |
| Claude Sonnet 4.6 | Premium | 47.2 | ~55 tok/s |
| Qwen 3.7 Max | Open-source | 46 | ~196 tok/s |
| MiniMax M3 (this page) | Open-source | 44.4 | ~113 tok/s |
| DeepSeek V4 Pro | Open-source | 44.3 | ~62 tok/s |
| GPT-5.3 Codex | Premium | 44.3 | ~106 tok/s |
| Kimi K2.6 | Open-source | 44.2 | ~43 tok/s |
| MiMo V2.5 Pro | Open-source | 42.2 | ~56 tok/s |
| Kimi K2.7 Code | Open-source | 41.9 | ~46 tok/s |
| DeepSeek V4 Flash | Open-source | 40.3 | ~106 tok/s |
| GLM-5.1 | Open-source | 40.2 | ~81 tok/s |
| Qwen 3.6 Max Preview | Open-source | 40 | ~46 tok/s |
| GPT-5.4 Mini | Premium | 40 | ~171 tok/s |
| Qwen 3.6 Plus | Open-source | 39.6 | ~53 tok/s |
| GLM-5 | Open-source | 39.5 | ~51 tok/s |
| Qwen 3.7 Plus | Open-source | 39 | ~52 tok/s |
| Kimi K2.5 | Open-source | 38.1 | ~51 tok/s |
| MiniMax M2.7 | Open-source | 38.1 | ~49 tok/s |
| Nemotron 3 Ultra | Open-source | 37.8 | ~204 tok/s |
| MiMo V2.5 | Open-source | 37.2 | ~88 tok/s |
| MiniMax M2.5 | Open-source | 33.7 | ~78 tok/s |
| Step 3.7 Flash | Open-source | 30.3 | ~407 tok/s |
| Step 3.5 Flash | Open-source | 26 | ~207 tok/s |
| Gemini 3.1 Flash Lite | Premium | 25 | ~300 tok/s |
| Claude Haiku 4.5 | Premium | 23.7 | ~103 tok/s |
| Kimi K3 | Open-source | Not yet scored | — |
| Kimi K2.7 Code HighSpeed | Open-source | Not yet scored | — |
| GLM-5.2 Fast | Open-source | Not yet scored | — |
| Inkling | Open-source | Not yet scored | — |
| Laguna S 2.1 | Open-source | Not yet scored | — |
| Gemini 3.6 Flash | Premium | Not yet scored | — |
| Gemini 3.5 Flash Lite | Premium | Not yet scored | — |
| Fugu Ultra | Premium | Not yet scored | — |
Frequently asked questions
MiniMax M3 or Qwen 3.7 Max?
Qwen 3.7 Max scores higher on the Intelligence Index (46 vs 44.4) at $1.875 blended per million tokens against MiniMax M3's $0.525. Default to MiniMax M3 and switch up when a task keeps stalling.
MiniMax M3 or DeepSeek V4 Flash?
DeepSeek V4 Flash is about 67% cheaper blended ($0.175 vs $0.525 per million tokens), scoring 40.3 on the Intelligence Index. Use DeepSeek V4 Flash for volume work and MiniMax M3 where its edge earns the difference.
How much does MiniMax M3 cost in Command Code?
$0.30 per million input tokens and $1.20 per million output tokens, with cache reads at $0.06. In an agent loop, cached context brings effective input to roughly $0.132 per million tokens.
What plan do I need for MiniMax M3?
MiniMax M3 is available on every plan, including Go at $1/mo.
How good is MiniMax M3 at coding?
MiniMax M3 scores 58.6 on the Coding Index — the Artificial Analysis sub-score most predictive of coding-agent performance — alongside 44.4 overall.
Does MiniMax M3 support image input and reasoning?
Yes — MiniMax M3 accepts image input alongside text. It supports reasoning.
Which Command Code model should I use?
Claude Fable 5 currently leads the lineup on the Intelligence Index (59.9). Grok 4.5 (53.8) leads the open-weights tier, available on every plan. For fast lookups, Step 3.7 Flash streams ~407 tok/s. There is no single right answer — switch per session with /model and let the task pick the model.
Can I mix MiniMax M3 with other models in a workflow?
Yes. Switch per session using /model. Common pattern: keep a default model and switch up for hard problems or down for quick lookups as the task calls for it.
Are open-source model prices fixed?
Open-source models are routed across multiple upstream providers for high availability. The price listed for each is the mean per-provider rate. Actual cost on a given request may vary slightly. The Usage page reflects the price charged.
Why is MiniMax M3 so cheap for a frontier model?
Open-source models are routed across multiple upstream providers; the listed price is the mean per-provider rate. The Usage page reflects the actual price charged.
Why no Intelligence Index for M3?
Public aggregate benchmarks have not yet been published for M3 in the current Intelligence Index format. The model is available and routed normally.
Is Command Code free to try?
The Go plan starts at $1/mo with $10 in LLM credits. It covers open-source models only. Pro at $15/mo unlocks premium models with $30 in LLM credits.
Does Command Code train on my code?
No. Command Code does not train on your code or store your code snippets. taste-1 data is stored locally in your project directory.
Where can I track my usage?
The Usage page in Studio shows per-request cost, token counts, and which model ran. Settings > Billing lets you change plans, buy credits, or enable auto top-up.
Does Command Code replace my editor?
No. Command Code is editor-agnostic — it runs as a CLI and works alongside any editor (Cursor, VS Code, Zed, JetBrains, Neovim, etc.).
Related reading
- MiniMax M2.7 in Command Codeend-to-end software engineering agent
- MiniMax M2.5 in Command Codecross-platform full-stack agentic dev
- DeepSeek V4 Pro in Command Codehybrid-attention long-context reasoning
- GPT-5.3 Codex in Command Codefrontier coding model
- Every model, one referenceThe docs list of all Command Code models with ids and context windows.
- Pricing, limits, and dealsThe canonical price table, running deals, and usage estimates.
Ship code that matches your taste
Command Code is the AI coding agent that continuously learns your taste. Start for $1.