MiniMax (Hailuo AI)
Listed at https://platform.minimaxi.com
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | MiniMax-01 Lightning: ¥0.2/M tokens, Turbo: ¥0.8/M, Pro: ¥2/M, M3: ¥1.5/M | per million tokens |
| Output | MiniMax-01 Lightning: ¥0.2/M tokens, Turbo: ¥0.8/M, Pro: ¥2/M, M3: ¥1.5/M | per million tokens |
🤖 Supported Models (10)
✨ Pros
- ✓MiniMax-01 series: Lightning/Turbo/Pro tiers, ¥0.2-2/M across the widest price range
- ✓MiniMax-M3: 2026 flagship MoE model with strong bilingual capability, coding near GPT-5 level
- ✓Full multimodal: text + vision (VL-01) + TTS + music gen + image gen
- ✓Hailuo AI consumer brand widely recognized, growing API ecosystem
- ✓Direct access from China, no proxy needed
- ✓From ¥0.2/M tokens, extremely competitive pricing
⚠️ Cons
- ×International recognition behind DeepSeek/Qwen
- ×English API documentation less comprehensive
- ×Some premium features (M3) require enterprise application
- ×Fast model iteration pace, version management challenging
- ×No native Function Calling support
🎯 Best For
China-based developers; multimodal applications (text+image+voice+music); cost-priority scenarios
💰 Pricing & Plans
| Model | Input ($/M tokens) | Output ($/M tokens) | Notes |
|---|---|---|---|
| MiniMax-M3 (≤512K, Standard) | $0.30 | $1.20 | Permanent 50% off intro price |
| MiniMax-M3 (>512K, Standard) | $0.60 | $2.40 | Long-context band, same 50% off |
| MiniMax-M3 (Priority tier) | $0.45 | $1.80 | Higher rate for guaranteed throughput |
| MiniMax-M2.7 | $0.30 | $1.20 | Mid-tier default, balanced cost/perf |
| MiniMax-M2.7-highspeed | $0.60 | $2.40 | Lower-latency variant |
| MiniMax-M2.5 / M2.1 / M2 (Legacy) | $0.30 | $1.20 | Discounted legacy line, still supported |
🔧 API & Developer Experience
- •API Style: OpenAI-compatible /v1/chat/completions endpoint — drop-in replacement for OpenAI SDKs, supports streaming (SSE), JSON mode, and system/user/assistant roles with max_tokens control.
- •SDK Availability: Official Python and Node SDKs (openai-compatible wrappers) plus community Go/Java SDKs; Anthropic and Gemini users migrate with minimal code changes.
- •Tool Use / Function Calling: Native function-calling support on M3 and M2.7 with structured JSON schema inputs; chains reliably across multi-turn agent loops, including parallel tool calls.
- •Long Context: Up to 1M-token context on M2 series with two-tier pricing (≤512K standard rate, >512K long-context band) — ideal for code repos, RAG corpora, and long-doc summarization.
- •Multimodal API Surface: Single API key covers text (LLM), vision (image-01, $0.0035/image), TTS (speech-2.8-hd/turbo), video (Hailuo-2.3, H3), and music generation — one billing account, one auth flow.
- •Billing & Trial: Pay-as-you-go standard Open Platform API keys billed by token; Token Plan subscriptions (Plus $22, Max $55, Ultra $132/month) bundle shared quotas across M3/M2.7/image/speech models.
🎬 Multimodal & Hailuo Video Ecosystem
MiniMax's distinctive edge is end-to-end multimodal generation under one API key: the same account that runs LLM inference also drives speech-2.8-hd/turbo TTS, image-01 vision synthesis, the Hailuo-2.3 and Hailuo-H3 video generation models (billed per second at $0.05–$0.13/sec depending on resolution), and Music-2.5+/3.0 generation. Hailuo-H3 added in mid-2026 is the flagship open general-purpose multimodal video model, with 768P at $0.08/s and 2K at $0.13/s, plus a $0.05/s regeneration path that re-renders low-res outputs to higher fidelity without re-billing the source prompt. Voice Design ($3/voice) and Rapid Voice Cloning ($1.5/voice) round out the audio stack. For teams building video-first applications—short-form content, marketing clips, AI hosts—this consolidated stack removes the multi-vendor integration tax that competitors force when pairing an LLM provider with a separate video/image vendor.
🌐 China Access & Latency
MiniMax is the inverse of U.S. providers on China access — its primary data center footprint is in Shanghai (operated by Shanghai Xiyu Technology / MiniMax AI), and the platform.MiniMax.io API endpoints are directly reachable from mainland China without VPN, proxy, or overseas routing. Beijing/Shanghai/Shenzhen clients typically see 30–80 ms latency on the LLM endpoint, comparable to any domestic cloud. The Hailuo AI consumer brand has high recognition in the China market, which simplifies enterprise procurement and invoicing. The trade-off is overseas latency: clients in North America and Europe route through China-based POPs and observe 180–260 ms round-trips on first token, with media-heavy endpoints (video, image) often slower due to cross-Pacific transfer. Developers building for global reach should pair MiniMax with a U.S./EU provider (OpenAI, Anthropic, or a gateway like Cloudflare AI Gateway) and route by region; developers shipping China-first products get a rare single-vendor stack with no geo-restrictions and no GFW risk.