MiniMax M3.1 on the official API is the model id MiniMax-M3.1-Flash. MiniMax’s Model Invocation and Model List pages list that name next to MiniMax-M3; they do not list a separate non-Flash MiniMax-M3.1 SKU. The 19 October 2026 product post is titled MiniMax M3.1 Flash and is the launch note for this line.
MiniMax M3.1 Flash as the API SKU
On the docs “Supported Models” table, MiniMax-M3.1-Flash is described as a frontier multimodal coding model with a 1,000,000-token context window and tunable thinking depth. MiniMax positions it for agentic reasoning, tool use, coding, and long-context work. Multimodal chat input covers text, images, and video.
Thinking is on by default and cannot be disabled: sending thinking: {"type": "disabled"} or effort: "none" returns HTTP 400 (model "MiniMax-M3.1-Flash" requires adaptive thinking). Depth is controlled with effort: low, medium, high, xhigh, or max (default max when omitted). On the Anthropic-compatible path the field is output_config.effort; on the OpenAI-compatible path it is reasoning_effort.
Endpoints documented by MiniMax:
- Anthropic-compatible (recommended):
https://api.minimax.io/anthropic - OpenAI-compatible:
https://api.minimax.io/v1
Call examples in the same guide use "model": "MiniMax-M3.1-Flash".
Official pricing
The Pricing Overview lists a flat pay-as-you-go table for MiniMax-M3.1-Flash (USD per million tokens):
| Model | Input | Output | Prompt caching read |
|---|---|---|---|
| MiniMax-M3.1-Flash | $0.15 | $0.60 | $0.015 |
Billing is by token count; MiniMax notes the token-to-character ratio varies by scenario. The blog repeats the same three rates and says agentic-task usage cost is about 70% lower than the predecessor, with online inference nearly 3× faster. Speed figures in the post are company-reported: everyday output can reach 200+ tokens per second, with median (P50) speeds consistently 170+ TPS.
Availability on the same post: MiniMax API, M Plan, and MiniMax Code; model weights will be released later. Desktop MiniMax Code download is linked at https://agent.minimax.io/download. The post also says a higher-capability Pro line will arrive later—treat that as a future product notice, not a current API id.
MiniMax-M3 still listed beside Flash
MiniMax-M3 remains a supported LLM with the same 1,000,000-token context window on the Model Invocation page. Pricing for M3 is separate and length/tiered on the Pricing Overview: Standard and Priority tables with permanent 50% off list rates, split at ≤512k vs >512k input tokens. Priority is described as 1.5× Standard and enabled with service_tier set to priority. Those M3 numbers are not the Flash flat rates above—do not mix them.
Migration note for older M2.x names
MiniMax’s Model List FAQ states that MiniMax M2.5 and M2.7 families stop for new users on 19 October 2026 and retire on 18 November 2026. For M Plan and Token Plan users, from 19 October 2026 requests to M2.5 and M2.7 are automatically routed to M3.1 Flash, while existing model names keep working for compatibility. Pay-as-you-go users on those older ids keep access until the November retirement date.
Company-reported benchmarks (not third-party audited here)
The launch post publishes several scores as MiniMax’s own results, including AutomationBench-AA standing, BankerToolBench 76.2, DeepSWE v1.1 73.7%, OSWorld V2 61.7, Agents’ Last Exam (ALE) 29.6, and CyberGym 82%. It also says Space Bunny (200+ trillion tokens processed anonymously) was an early version of M3.1 Flash. These are vendor claims, not independently re-run benchmark results.
→ Model rankings: Compare text models on the AI Rank model leaderboard.
Arena snapshot for MiniMax-M3 only
On AI Rank’s Arena overall feed (source publish date 2026-10-15), minimax-m3 appears (rank 96 in the snapshot used for this piece). MiniMax-M3.1-Flash was not present under that name in the same list. Do not read the M3 Arena row as a Flash score.
Sources
- MiniMax M3.1 Flash announcement (2026-10-19)
- Model Invocation / text generation
- Pricing Overview
- Model List FAQ
- AI Rank models
Keep exploring: see live model rankings on AI Rank.
Or 📱 get the AI Rank app to follow them on your phone.
Prepared with automated editorial assistance and checked against MiniMax docs and the 19 Oct 2026 announcement on October 24, 2026. We did not run our own latency or quality tests for this article.