MiniMax-M3.1-Flash

# Fix list: MiniMax-M3.1-Flash

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/minimax-m3-1-flash, the October 2026 research run, assessed 24 October 2026. Grade D, 50.5 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on MiniMax-M3.1-Flash: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Reach, 40 out of 100, up to 15 more on the total

Why it scored 40: Memory earns nothing. MiniMax's only build is its 250 GB NVFP4 and MXFP8 release, about 300 GB with the fifth added, above 256 GB, and no runtime library has a local build. Runtimes earn 5 for llama.cpp, whose release b11861 of 24 October 2026 runs it; vLLM needed a nightly image and SGLang an unmerged pull request, Transformers support is still in progress per the card, Ollama is cloud only and LM Studio has no entry. Builds earn 10 for one publisher precision. Terms earn 10, since commercial use is free with attribution and a notice below $20M of yearly revenue. Download earns 10, with no gate. Hosted earns 5, since MiniMax's API serves it.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reach):

Whether an agent can run it, and where.

- 0 to 30, memory. The smallest build from the publisher or a runtime's own library that keeps 4-bit precision or better, at its file size plus a fifth for the runtime and a working context. 30 for 16 GB or less, 25 for 24 GB, 20 for 32 GB, 15 for 64 GB, 10 for what fits our reference machine (the GPU share of 128 GB of unified memory), 5 for 256 GB, 0 above.
- 0 to 25, runtimes. 5 for each of llama.cpp, Ollama, vLLM, SGLang, LM Studio and Hugging Face Transformers that runs the architecture in a released version, from the runtime's own code, library or documentation. A publisher's fork or an unreleased branch doesn't count. 25 at most.
- 0 to 15, builds from the publisher. 15 for quantised builds at two precisions or more (FP8, 4-bit, GGUF), 10 for one, 0 when only other people's quantisations exist.
- 0 to 15, terms of use. 15 for an OSI-approved licence with nothing added, 10 for a custom licence that lets anyone use it commercially for free below a threshold or under a use policy, 5 for research or non-commercial use only, 0 for no licence or unclear terms.
- 0 to 10, download. 10 when the files download with no account and no terms to accept, 5 for a click-through that needs an account, 0 when a person approves each request.
- 5, hosted too. The publisher's own API serves it, or at least two inference providers do, so an agent without the hardware can still call it.

When our reference-machine runs start, speed on that machine (tokens per second and time to first token at the build that fits) joins Reach.

## 2. Risk, 35 out of 100, up to 13 more on the total

Why it scored 35: File safety earns 15, safetensors only, less 10 because the SGLang route the card links passes trust_remote_code for the repository's configuration and processor code. Provenance earns the full 15, files in the organisation MiniMax links to with a SHA-256 for each shard. Safety evaluation and agent guidance earn nothing; the launch post's security testing measured offensive capability, and the template's root role isn't documented as a trained hierarchy. Licence risk earns 5, after the military-use ban and the $20M threshold. The security programme earns nothing, with no security.txt, disclosure policy or advisories.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-risk):

What could go wrong when you run it, and what the publisher has done about it. A higher score means less risk.

- 0 to 25, file safety. 25 for weights in safetensors or GGUF only, 15 when pickle files (.bin, .pt, .pth) sit beside them, 0 for pickle only. 10 off when the documented way to load it or to build its prompts runs code from the repository (`trust_remote_code`, or the publisher's scripts in place of a template).
- 0 to 15, provenance. 10 when the files come from the publisher's own organisation, the one its site or docs link to, and 5 for a checksum on every file.
- 0 to 20, safety evaluation. 20 when results are published and someone other than the publisher ran them or can rerun them, 10 for the publisher's own results, 0 for none.
- 0 to 15, guidance for agent use. 15 for documented prompt-injection guidance or an instruction hierarchy the model was trained to follow, 10 for a statement of what its trained refusals cover and the safeguards to add around it, 5 for general cautions, 0 for none.
- 0 to 15, licence risk. 15 to start, 5 off for each of these, a use policy or field-of-use limits beyond the law, terms the licensor can change or revoke, a revenue or user threshold, and a licence that changed without a dated note or whose link is broken. Not below 0.
- 0 to 10, a security programme. 5 for a way to report a flaw (security.txt, a disclosure policy or a security contact), 5 for advisories handled in public.

When our reference-machine runs start, measured robustness (prompt injection with AgentDojo, over-refusal with XSTest) joins Risk.

## 3. Capability, 64 out of 100, up to 12.6 more on the total

Why it scored 64: From 6 independent benchmark results in 4 areas, each scored as its share of the way from the benchmark's floor to the best result any model has on the same leaderboard. By area, Long context 90, Coding 80, Reasoning and knowledge 76, Tool calling and agent tasks 40. Publisher-submitted results count at half weight, and numbers that appear only on the publisher's card don't count.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-capability):

How well the model does on the benchmarks that matter to an agent. We don't average a publisher's own numbers into a grade. A result counts in full when we ran it on our reference machine, or when an independent party ran it and published the method (the Berkeley Function Calling Leaderboard, Artificial Analysis, Epoch AI's Benchmarking Hub, the SWE-bench and Terminal-Bench leaderboards, the Open ASR Leaderboard and others that run models themselves). A result the publisher submitted to such a leaderboard counts at half weight. A number that appears only on the publisher's card is shown on the listing as the publisher's claim and doesn't count.

Each result is scored as its share of the way from the benchmark's floor (chance for multiple choice, zero for a pass rate) to the best result any model, open or closed, has on the same leaderboard and setting on the day we read it. A model that matches the best model anywhere scores 100 on that benchmark. For an error rate such as word error rate, lower is better and the scale runs the other way, and where the leaderboard gives no floor the rate is read against the best model's, so twice the best model's error rate scores 50. Speed as a multiple of real time (RTFx) runs from about 1 to over 20,000, so it's read on a log scale from real time to the fastest model. On the MTEB leaderboard, where most models score between 0.5 and 0.75, the floor is the median of every model the same tab lists, and its results count at half weight because MTEB doesn't say who ran them and most are submitted with the model. An arena rating (Elo) isn't a ratio, so its floor is 400 points below the best model on that arena, the gap at which the better model wins about ten games in eleven. Where a model has results at several settings (reasoning effort, say), its best one counts, because that's the setting an agent would run.

The results group into areas, each weighted within the category.

- Tool calling and agent tasks, 35 (BFCL, τ²-bench, Terminal-Bench and agent suites like them)
- Coding, 20 (SWE-bench Verified, LiveCodeBench, Aider Polyglot)
- Reasoning and knowledge, 20 (GPQA Diamond, MMLU-Pro, Humanity's Last Exam, composite indexes that publish their components)
- Instruction following, 15 (IFEval, IFBench)
- Long context, 10 (RULER, LongBench and long-context retrieval tests)

A speech model is read on its own areas, recognition accuracy (word error rate on the Open ASR Leaderboard's sets, 60) and speed (real-time factor, 40) for speech to text, and intelligibility and speaker similarity (Seed-TTS-eval, the Open TTS Leaderboard, 70) and speed (30) for text to speech. A publisher's own converted copy of the same weights (Qwen's -hf builds, say) counts as the same checkpoint. An embedding model is read on retrieval (60) and the other MTEB task types, classification, clustering, reranking and similarity among them (40), from the MTEB leaderboard. A vision model adds an image area (MMMU, ChartQA, DocVQA or OCRBench) at 20, and the others shrink to fit. A document parser, a model that turns page images into structured text, is read on the five things ParseBench measures, tables, charts, content faithfulness, semantic formatting and visual grounding, 20 each, from ParseBench's leaderboard or another independent document benchmark that publishes the same dimensions (since 22 October 2026). One leaderboard gives all five at once, so a parser's confidence counts independent sources, and a row the publisher submitted (or that the leaderboard copied from the publisher's card) counts at half weight with low confidence. LlamaIndex, which runs ParseBench, also sells LlamaParse, a parser the same benchmark ranks, and the listing says so.

Capability is the weighted mean of the areas with results, and needs results in at least two areas (one for a speech, embedding or document-parsing model). With less than that it's pending for that listing, its weight is shared across the other four, and the total is capped at 61, so a model nobody has benchmarked independently can't reach B, and every model with results ranked B or better stays above it. The listing says so. Until 21 October 2026 the cap was 69, which put models nobody had benchmarked above ones that had been.

## 4. Documentation and upkeep, 41 out of 100, up to 5.9 more on the total

Why it scored 41: The model card earns 11, missing limitations, languages and a knowledge cut-off. Training data earns nothing, as it isn't described. The last change earns 20, the release commit of 23 October 2026. Responsiveness earns nothing, with no MiniMax reply in five discussions, though the release is a day old. Version policy earns 5 for a first release in its own repository. File health earns 5, for the tokenizer's 40,960,000-token model_max_length against the 1M context, the Transformers library tag against the card, and an unanswered vLLM streaming report.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-upkeep):

- 0 to 20, the model card. Intended use, limitations, architecture, parameters, context length, languages and knowledge cut-off. 20 for all seven, 3 off for each that's missing.
- 0 to 20, training data. 20 for the data published or listed with sources and sizes, 15 for a data card that names the sources or the mix, the cut-off and the filtering, 10 for a general description with a cut-off, 5 without one, 0 for none.
- 0 to 20, the last change to the files, the template or the reference code. 20 within 30 days, 15 within 90, 10 within 180, 5 within a year, 0 older.
- 0 to 20, responsiveness. The publisher's replies to the 40 most recent questions and reports in the repository's discussions or issues. 20 for most, 10 for some, 5 for a few, 0 for none.
- 0 to 10, version policy. 10 for earlier versions kept in their own repositories with a dated note when one is replaced, 5 for earlier versions kept without notes or a first release, 0 when files were replaced in place without a note.
- 0 to 10, file health. 10 when config, tokenizer, template and card agree and no report of a broken file waits without a reply, 5 when one does, 0 when the files contradict the card.

## 5. Agent interface, 70 out of 100, up to 3 more on the total

Why it scored 70: The chat template earns 20 for chat_template.jinja in the repository. The tool-calling format earns 10, since only the template shows the format and the parsers come from vLLM's and SGLang's own docs. Trained tool calling earns 20, with several invoke blocks in one call. Structured output earns nothing, as it isn't documented. Reasoning control earns 10, with reasoning_effort per request on the card and the reasoning in mm:think tags. Context earns 10 for 1,048,576 tokens with no position scaling.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-interface):

What a model and a developer need to talk to it. Speech, image and other models with no chat interface read the tool-calling and structured-output lines for what the model takes and returns (voices, timestamps, streaming, typed outputs), and the note says how.

- 0 to 20, the chat template. 20 for a template in the repository (Jinja, in the tokenizer config or its own file), 10 when the prompt format is only given as code you run.
- 0 to 25, the tool-calling format. 25 when the publisher documents how tools are declared and how calls come back, with the parser to use in a named runtime, 10 when only the template or a runtime's own recipe says it.
- 0 to 20, tool calling the model was trained for. 20 with several calls in one turn, 10 for one call at a time.
- 0 to 15, structured output. 15 when JSON or JSON Schema output is documented for the model with how to ask for it, 5 when the card claims it without saying how.
- 10, reasoning control. Thinking that can be turned off or budgeted per request, with where the reasoning text comes back.
- 0 to 10, context. 10 for 128K tokens or more trained, 5 for 32K to 127K. Context reached only by scaling the position encoding counts one band down.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- Runtimes score 5 because vLLM and SGLang needed pre-release builds on 24 October 2026. When vLLM 0.33.0 ships, or SGLang merges support, each adds 5, and Reach rises to 45 or 50
- Capability uses Artificial Analysis's and Vals AI's runs of MiniMax's API. MiniMax doesn't say the API serves these same weights, though the card's benchmark table matches the API launch post and says its results are for the official weights; Artificial Analysis still lists the model as proprietary
- No sameModel set for minimax-api, for the same reason. MiniMax's release notes call M3.1 Flash an open-weight model, which the founder may read as enough
- Agent guidance is 0. The template's root role is labelled high priority over system and developer messages; if that's read as a trained instruction hierarchy the line gives 15 and Risk 50
- Terms read section 2 of the licence as allowing free commercial use below $20M with attribution and a notice, although the grant says non-commercial purposes; a reading of the grant alone would give 5
- Responsiveness is 0 on five discussions in the release's first day; a recount in a few weeks may move it
- unchecked, GitHub issues; MiniMax has no public repository for M3.1 Flash, and github.com web pages answer this session with 403

## Weaknesses

- No released vLLM or SGLang ran it on 24 October 2026; only llama.cpp b11861 from that day did
- About 300 GB with a fifth added, more than any 128 GB or 256 GB machine
- Commercial use needs a Built with line and a notice to MiniMax, and authorisation above $20M yearly revenue
- No safety evaluation, agent guidance or security contact, alongside strong claimed offensive security capability
- The card omits limitations, languages, training data and a knowledge cut-off, and no discussion has a MiniMax reply

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Check the use case first; commercial products must show "Built with MiniMax M3.1-Flash" and notify MiniMax once, or get written authorisation above $20M yearly revenue
- On 8 H100 or H200 GPUs use vLLM's nightly image vllm/vllm-openai:minimax-m31-flash with --tensor-parallel-size 8 --block-size 128 --kv-cache-dtype fp8 --moe-backend marlin --max-model-len 262144
- Pass --reasoning-parser minimax_m3 --tool-call-parser minimax_m3 --enable-auto-tool-choice in vLLM, or minimax-m3 for both in SGLang with --trust-remote-code
- --block-size 128 is mandatory in vLLM, since MSA's sparse block size is 128
- Set effort per request with chat_template_kwargs reasoning_effort (low, medium, high, xhigh, max); sample at temperature 1.0 and top_p 0.95
- On Blackwell or MI355X, 4 GPUs serve it with the DSpark draft from dspark/; vLLM's Hopper command leaves DSpark off and turns video input off
- For GGUF builds by others, use llama.cpp b11861 or later; earlier builds mis-scale the NVFP4 experts and produce garbage
- Without the hardware, call MiniMax-M3.1-Flash on MiniMax's API (minimax-api), whose thinking defaults to max effort

## What the review panel asked for

- a security contact
- safety evaluation results
- a changelog for the weights
- replies in discussions
- Document the tool format
- Fix model_max_length
- Release a smaller build
- Ship a released vLLM

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.