MiniMax-M2.7

# Fix list: MiniMax-M2.7

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/minimax-m2-7, the October 2026 research run, assessed 21 October 2026. Grade C, 56.1 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-M2.7: 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. Risk, 30 out of 100, up to 14 more on the total

Why it scored 30: File safety earns 15, safetensors only with no pickle files, less 10 because every documented loading route passes trust_remote_code for the repository's Python. 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, as MiniMax publishes neither. Licence risk earns nothing (read 21 October 2026). The Prohibited Uses appendix goes beyond the law, MiniMax swapped a commercial licence for a non-commercial one two days after release, and neither that swap nor the amendment three days later carries a dated note. The security programme earns nothing, with no security.txt, disclosure policy, SECURITY.md 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.

## 2. Capability, 71 out of 100, up to 10.1 more on the total

Why it scored 71: From 16 independent benchmark results in 5 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, Instruction following 91, Long context 88, Coding 80, Reasoning and knowledge 70, Tool calling and agent tasks 53. 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.

Capability is the weighted mean of the areas with results, and needs results in at least two areas (one for a speech or embedding 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.

## 3. Reach, 60 out of 100, up to 10 more on the total

Why it scored 60: Memory earns 5. MiniMax only publishes its 230 GB FP8 release, and the smallest build of 4-bit or better from a runtime's own library is LM Studio's Q4_K_M GGUF at 138.3 GB, about 166 GB with the fifth added, so it needs more than our 107.4 GB machine and fits within 256 GB. Runtimes earn the full 25, with model code in released vLLM, SGLang, llama.cpp and Transformers and an LM Studio catalogue entry; Ollama only serves it from its cloud. Builds earn 10 for one publisher precision, the FP8 release. Terms earn 5, since the licence today allows only personal, research and non-profit use without MiniMax's written authorisation. Download earns 10, with no gate or account. Hosted earns 5, since Novita, Featherless AI and DeepInfra serve it through Hugging Face, while MiniMax's own API is closed to new users and retires it on 18 November 2026.

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.

## 4. Documentation and upkeep, 27 out of 100, up to 7.3 more on the total

Why it scored 27: The model card earns 2, stating only intended use and leaving out limitations, architecture, parameters, context, natural languages and a cut-off. Training data earns nothing, since only the training process is described. The last change earns 5, a config edit on 20 April 2026, within a year. Responsiveness earns 10, with a MiniMax reply in 8 of 36 discussions, none since June. Version policy earns 5, since earlier and later models sit in their own repositories without dated notes. File health earns 5, because a report that it doesn't load in Transformers v5 waits without a reply.

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, 75 out of 100, up to 2.5 more on the total

Why it scored 75: The chat template earns 20 for chat_template.jinja in the repository. The tool-calling format earns 25, as MiniMax's own guide documents how tools go in and how calls come back, gives a parsing function, and its deployment guides name the vLLM and SGLang parsers. Trained tool calling earns 20, with several invoke blocks in one call. Structured output earns nothing, since JSON output isn't documented or claimed. Reasoning control earns nothing, because thinking is always on with no documented way to turn it off or budget it. Context earns 10 for 204,800 tokens in the config 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.

- Whether the licence swap of 11 April 2026, from a commercial licence to a non-commercial one two days after release with no note, is also a negative event is the founder's call; none is applied, as for minimax-m2-5
- Whether people who downloaded the files between 9 and 11 April 2026 keep the earlier MiniMax Model License is a legal question we didn't settle; terms are read on the licence as it is today
- Responsiveness reads 8 of 36 as some (10); a reader who sees it as a few would give 5
- Unchecked, replies on the MiniMax-AI/MiniMax-M2.7 GitHub issues; github.com answered 403 to this session, so responsiveness rests on the Hugging Face discussions alone
- Builds read as one precision (10) for MiniMax's native FP8 release, as for the other native FP8 releases; a stricter reading would give 0
- No sameModel set; MiniMax's API still serves M2.7 to existing users until 18 November 2026 and its model list already marks it retired
- The model doesn't fit the 107.4 GB reference machine at any publisher or runtime-library build (the smallest, LM Studio's Q4_K_M, needs about 166 GB), so our own runs won't add speed or robustness

## Weaknesses

- Commercial use needs MiniMax's written authorisation since a licence swap on 11 April 2026, made without a note
- 230 GB in FP8, and about 166 GB for the smallest 4-bit library build, more than a 128 GB machine
- Thinking is always on, with no documented way to turn it off or budget it
- No safety evaluation, agent guidance or security contact, and every loading route runs repository code
- No MiniMax reply in discussions since June 2026, and the card leaves out limitations, architecture, parameters and a cut-off

## 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 is personal, research or non-profit before starting it; anything commercial needs MiniMax's written authorisation, and commercial users must show Built with MiniMax M2.7
- Start it with vLLM on 4 GPUs of 96 GB, SAFETENSORS_FAST_GPU=1 vllm serve MiniMaxAI/MiniMax-M2.7 --trust-remote-code --tensor-parallel-size 4 --enable-auto-tool-choice --tool-call-parser minimax_m2 --reasoning-parser minimax_m2_append_think
- In SGLang use --tool-call-parser minimax-m2 and --reasoning-parser minimax-append-think, with --trust-remote-code
- MiniMax's guides budget 220 GB for the weights and 240 GB per 1M tokens of KV cache, with 196K the most per sequence
- On one large-memory machine, LM Studio's catalogue entry minimax/minimax-m2.7 lists a 138 GB Q4_K_M GGUF and states 138 GB minimum memory
- Sample with temperature 1.0, top_p 0.95 and top_k 40, as the card recommends, and budget output tokens for thinking on every turn
- Pin Transformers 4.57.1 as MiniMax's guide does; a report that it fails on Transformers v5 has no answer
- Without the hardware, call it through Novita, Featherless AI or DeepInfra; MiniMax's own API retires M2.7 on 18 November 2026

## What the review panel asked for

- record licence changes on the card
- published commercial terms
- a way to turn thinking off
- dated licence change notes
- a security contact

## 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.