# MiniMax API (slim)

> MiniMax's API for its M-series text models, M3.1 Flash and M3, in the Anthropic Messages, OpenAI Chat Completions and OpenAI Responses formats, with 1M-token context and automatic prompt caching.

- Full: https://www.anchorterminal.com/tools/minimax-api.md (~10,900 tokens) · this version ~2,280 tokens · JSON https://www.anchorterminal.com/tools/minimax-api.json · canonical https://www.anchorterminal.com/tools/minimax-api
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-19 · pricing https://www.anchorterminal.com/tools/minimax-api#pricing · reviews https://www.anchorterminal.com/tools/minimax-api#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 55.2/100 · rank #751 of 1175 · #19 in Model APIs & inference · not agent-ready · confidence medium**


Assessment: M3.1 Flash costs $0.15 in and $0.60 out per million tokens, behind Anthropic and OpenAI request formats with OpenAPI files and 1M-token context. The terms let MiniMax use inputs and outputs to improve its services with no opt-out found, and the status page logged 23 short LLM incidents in the week to 19 October 2026.

## Facts

- Kind: Model API · vendor: MiniMax · category: Model APIs & inference · legal entity: Nanonoble Pte. Ltd. · provenance 80/100
- Endpoint: `https://api.minimax.io` (HTTP)
- Auth: API key · pricing: Pay per use · x402: no · licence: Proprietary API under MiniMax's platform terms. MiniMax-M3 weights are on Hugging Face under the MiniMax community licence
- Probe metrics: not measured yet (probes haven't run)
- Models: MiniMax-M3.1-Flash and MiniMax-M3. Legacy MiniMax-M2.7, M2.7-highspeed, M2.5 and M2.5-highspeed closed to new users on 19 October 2026 and retire on 18 November 2026. M2.1, M2 and M2-her are retired
- Request formats: Anthropic Messages at api.minimax.io/anthropic (recommended), OpenAI Chat Completions and OpenAI Responses at api.minimax.io/v1, and MiniMax's own /v1/text/chatcompletion_v2. OpenAPI 3.1 files for each
- Context and output: 1,000,000-token context on both models. Up to 524,288 output tokens, 131,072 recommended. MiniMax's M3 page says 1M with a guaranteed minimum of 512K
- Thinking: M3.1 Flash always thinks. `effort` takes low, medium, high, xhigh or max and defaults to max, and sending `disabled` returns 400. M3 thinks only when `thinking` is set to adaptive
- Inputs: Text, images (up to 10 MB) and video (up to 50 MB, or 512 MB through the Files API) on M3.1 Flash and M3. Request bodies up to 64 MB
- Rate limits: 200 requests and 10,000,000 tokens a minute for M3.1 Flash and M3
- Free tier: None found
- Trains on API data: Unclear. The terms (effective 30 March 2026) let MiniMax use input and generated content to develop and improve its services. The privacy policy says personal data isn't used for training to profile or target consumers. No opt-out found
- Data location: A data centre in the United States, per the privacy policy updated 30 March 2026
- Batch and structured output: No batch API and no JSON Schema output mode documented. The Anthropic format ignores `stop_sequences` and `top_k`
- M Plan: Go $22, Explore $55 and Build $132 a month, with Explore at 3 times and Build at 7.5 times Go's usage. M3.1 Flash on all three and H3 video on Explore and Build, in MiniMax Code or any Anthropic- or OpenAI-compatible tool through a Subscription Key. Usage resets in 5-hour and weekly windows. MiniMax recommends pay as you go for production
- Open weights: MiniMax-M3 (about 428B parameters, about 23B active) on Hugging Face under the MiniMax community licence since 2 June 2026. M3.1 Flash weights are promised for later, with no repository on 19 October 2026
- Vendor benchmarks: Vendor-reported for M3.1 Flash in MiniMax's launch table. Automation-Bench v1.0.6 62.2 (62.8 at max effort on its chart), Agents' Last Exam 29.6, DeepSWE v1.1 73.7, BabyVision with tools 91.0, OSWorld 2.0 61.7, CyberGym 82.2. Not our measurement
- Status history: 23 'Elevated errors on Large Language Models' incidents from 13 to 19 October 2026, each resolved within 1 to 41 minutes. Earlier history wasn't readable
- Models ($/1M in/out): `MiniMax-M3.1-Flash` $0.15/$0.60; `MiniMax-M3` $0.30/$1.20
- Prices: web_search server tool (beta) $10 per 1,000 requests; M Plan Go $22 per month (plan); M Plan Explore $55 per month (plan); M Plan Build $132 per month (plan)
- 2026-10-19 Notice: MiniMax-M3.1-Flash released at $0.15 in and $0.60 out per 1M tokens, and the MiniMax-M3.1-Flash-Preview name now serves it
- 2026-10-19 Rename: MiniMax-M2.5 and M2.7 families closed to new users, and M Plan and Token Plan requests to them served by M3.1 Flash under the old names
- 2026-11-18 Shutdown: MiniMax-M2.5, M2.5-highspeed, M2.7 and M2.7-highspeed retire, replacement MiniMax-M3.1-Flash

- Scores: Reliability 75, Performance pending, Schema & documentation 82, Agent ergonomics 59, Security & auth 30, Payments & pricing 20, Task success pending, Maintenance & community 45, Transparency & trust 64 · total over the 7 assessed categories
- Why: Reliability, Atlassian Statuspage at status.minimax.io with a Large Language Models component (20). · Schema & documentation, OpenAPI 3.1 files for the Anthropic Messages, OpenAI Chat Completions and OpenAI Responses endpoints, each listing MiniMax-M3.1-Flash and Mi… · Agent ergonomics, Model reading of the checklist (tool use, structured output, caching, context, batch, SDKs, errors). · Security & auth, Model reading (credential, training, retention, operator visibility, programme). · Payments & pricing, No x402, MPP or L402 (0). · Maintenance & community, Model reading. · Transparency & trust, Closed service with clear terms under Singapore law and SIAC arbitration.
- Sources: 24, open questions: 7, both in the full twin
- Capabilities: inference.llm, inference.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/minimax-api.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/minimax-api.svg` or a link to https://www.anchorterminal.com/tools/minimax-api from a page on minimax.io or one of its subdomains, or the README of github.com/MiniMax-AI/MiniMax-M3, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Use the Anthropic-format base URL `https://api.minimax.io/anthropic` for tool calls. The OpenAI-format schema has no `tool_choice` or `response_format`
2. Set `output_config.effort` (or `reasoning_effort`) below `max` for routine calls. M3.1 Flash defaults to max, can't turn thinking off, and counts thinking against `max_tokens`
3. Don't rely on `stop_sequences` or `top_k`. The Anthropic-format endpoint ignores them
4. Back off on 429 and 529 yourself. No Retry-After header is documented
5. Use a pay-as-you-go key for production. A Subscription Key draws on M Plan's 5-hour and weekly windows and can't stand in for an API key

## Connect

```bash
curl https://api.minimax.io/anthropic/v1/messages \
  -H "Authorization: Bearer $MINIMAX_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"MiniMax-M3.1-Flash","output_config":{"effort":"low"},"max_tokens":4096,"messages":[{"role":"user","content":"hello"}]}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| GroqCloud | BB | 74.5 | inference.llm, inference.open-weights | https://www.anchorterminal.com/tools/groq.min.md |
| Fireworks AI Inference | BB | 73.5 | inference.llm, inference.open-weights | https://www.anchorterminal.com/tools/fireworks-inference.min.md |
| Mistral AI API | BB | 71.2 | inference.llm, inference.open-weights | https://www.anchorterminal.com/tools/mistral-api.min.md |
| Cerebras Inference | B | 68.4 | inference.llm, inference.open-weights | https://www.anchorterminal.com/tools/cerebras-inference.min.md |
| Cloudflare Workers AI | B | 68.3 | inference.llm, inference.open-weights | https://www.anchorterminal.com/tools/cloudflare-workers-ai.min.md |


## Panel reviews (4, average 2.8/5, desk reviews from public material, no calls made)

- ★★☆☆☆ Thirty days' notice, and old model names already serve M3.1 Flash (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★★☆ $2.10 per thousand mid-sized calls, plus thinking tokens (Ledger, Cost analyst, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★☆☆ Eight typed errors on one format, only a 200 on the others (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ No key scopes, and inputs may improve the service (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-19; 4 upheld, 0 corrected, 0 rejected): All four panel reviews of the MiniMax API match the dossier, and none needed correction. Ratings run from 2 to 4 because the lenses weigh different facts. Ledger finds M3.1 Flash costs $2.10 per 1,000 mid-sized calls before thinking tokens. Warden and Keel mark down terms that let MiniMax use inputs with no opt-out found, and 30 days' notice for the M2.5 and M2.7 retirement. A reader should check the data terms before sending anything they can't share with MiniMax.