China AI Hub
API / minimax
minimax API
Provider: minimax · Type: official
minimax is an API by minimax, served at https://api.minimax.io/anthropic for official models, exposing MiniMax-M2.7-Highspeed, MiniMax-M2.7, MiniMax-M3.
| Key fact | Value |
|---|---|
| Endpoint | https://api.minimax.io/anthropic |
| Authentication | API key (Bearer-style, passed via ANTHROPIC_API_KEY), managed in the platform console |
| Streaming | Yes |
| Function calling | Yes |
| Tool calling | Yes |
| Structured output | Unknown |
| Vision | Yes |
| Audio | Unknown |
| Regions | international, china |
| Cloud providers | MiniMax Platform |
| Documentation | https://platform.minimax.io/docs |
Context limits
| Model | Input limit | Output limit |
|---|---|---|
| minimax-m3 | 1,048,576 | — |
| minimax-m2.7 | 204,800 | — |
| minimax-m2.7-highspeed | 204,800 | — |
What it is. The MiniMax API serves the M-series through Anthropic-compatible and OpenAI-compatible endpoints, with international (api.minimax.io) and China (api.minimax.cn) bases. Why it matters. It is the primary way to access MiniMax-M3’s 1M-token context and multimodal input at the database’s lowest flagship price ($0.30/$1.20 per 1M tokens). Key characteristics. Anthropic-compatible endpoint with ANTHROPIC_API_KEY-style auth; streaming with thinking/text deltas; service_tier (standard | priority, priority = 1.5x price); automatic prompt caching. What a professional should know. Rate limits are published on a JS-rendered page (not extractable), and structured output is not publicly documented — both are known limitations to verify before production use.
The MiniMax API serves the M-series through Anthropic-compatible and OpenAI-compatible endpoints:
international base https://api.minimax.io/anthropic (China: https://api.minimax.cn/anthropic), plus
OpenAI Chat Completions and Responses endpoints and legacy MiniMax-native endpoints.
Streaming is supported (thinking streamed as thinking_delta, text as text_delta). Key parameters: max_tokens, temperature (0-2, recommended 1.0), top_p, tools/tool_choice, service_tier (standard | priority, priority = 1.5x price). M3 supports the thinking parameter (disabled by default, adaptive to enable) and multimodal Anthropic content blocks (image/video via URL, base64 or mm_file://{file_id}). Prompt caching is automatic with explicit cache_control support.
Why it matters
The MiniMax API matters as the Anthropic-compatibility play: by exposing an ANTHROPIC_API_KEY-style endpoint, it lets teams already on the Claude ecosystem switch models with minimal code change — a deliberate contrast to DeepSeek’s OpenAI-first surface. The cost trade-off is the service_tier structure (priority = 1.5x) layered on top of already-low M3 pricing, plus automatic prompt caching that lowers repeat-query cost. China AI Hub analysis indicates the API’s structural role is the value-compatible option: Anthropic-style integration at the database’s cheapest flagship price, with the gaps being structured output and published rate limits.
Labels used above: Official fact (from MiniMax’s Anthropic-compatible API and pricing docs), Vendor-reported claim (pricing and capability statements by MiniMax), and China AI Hub analysis (our synthesis, always introduced as such).
Sources
Data interpretation
| Field | Value | Evidence type |
|---|---|---|
| Endpoint | https://api.minimax.io/anthropic | Official |
| Authentication | API key (Bearer-style, passed via ANTHROPIC_API_KEY), managed in the platform console | Official |
| Capabilities | Streaming, Function calling, Tool calling, Vision | Official |
| Models served | 3 models in the database | China AI Hub analysis |
Evidence types: Official = vendor documentation or official pages. China AI Hub analysis = derived from the database itself. See the sourcing policy.
What is the minimax API endpoint?
The minimax API endpoint is https://api.minimax.io/anthropic.
How do you authenticate with the minimax API?
API key (Bearer-style, passed via ANTHROPIC_API_KEY), managed in the platform console.
What are the context limits of the minimax API?
minimax-m3: 1,048,576 in / — out; minimax-m2.7: 204,800 in / — out; minimax-m2.7-highspeed: 204,800 in / — out.
Does the minimax API support function calling?
Yes — the minimax API supports function calling.
Where does China AI Hub get its minimax API data?
From 2 sources, last verified 2026-09-20.