Amanai Docs
Chat
Chat model catalog, capabilities, reasoning controls, and compatible request formats.
Overview
Chat models share one authenticated catalog and work through OpenAI-compatible or Anthropic-compatible request formats. Change the model field without changing your account, base URL, or API key.
Dedicated documentation. Image, video, and speech models are in Media; live WebRTC sessions are in Realtime; vectors and ranking are in Embeddings and Rerank. Typed decisions use the separate System One endpoint.
Chat endpoints
| Endpoint | Compatibility | Use when |
|---|---|---|
POST /v1/chat/completions | OpenAI Chat Completions | Messages, streaming, tools, and broad SDK compatibility |
POST /v1/responses | OpenAI Responses | Responses SDK, structured output, and agent-style workflows |
POST /v1/messages | Anthropic Messages | Anthropic SDK and Claude Code-compatible clients |
Model catalog
The authenticated GET /v1/models response is authoritative for the models available to your key. The page starts from the latest promoted snapshot and refreshes from the public catalog outside preview mode.
curl https://api.amanai.sh/v1/models -H "Authorization: Bearer $AMANAI_API_KEY" 
GPT
| Model | Description |
|---|---|
GPT 5.5 amanai/gpt-5.5 | OpenAI GPT-5.5 — a new class of intelligence for coding and professional work; 1.05M context. |
GPT 6 Astra amanai/gpt-6-astra | OpenAI GPT-6 Astra — advanced reasoning model for complex professional work; 1.05M context, 128k output. |
GPT 5.6 Sol amanai/gpt-5.6-sol | OpenAI GPT-5.6 Sol — frontier model for complex reasoning and coding; 1.05M context, 128k output. |
GPT 5.6 Terra amanai/gpt-5.6-terra | OpenAI GPT-5.6 Terra — balances intelligence and cost (mini tier); 1.05M context. |
GPT 5.6 Luna amanai/gpt-5.6-luna | OpenAI GPT-5.6 Luna — cost-sensitive, high-volume workloads (nano tier); 1.05M context. |
Claude
| Model | Description |
|---|---|
Claude Fable 5.1 amanai/claude-fable-5.1 | Anthropic Claude Fable 5.1 — latest Fable tier for coding and agentic work; 1M context, 128k output. |
Claude Fable 5 amanai/claude-fable-5 | Anthropic Claude Fable 5 — newest Claude tier for coding and agentic work; 1M context, 128k output. |
Claude Opus 5 amanai/claude-opus-5 | Anthropic Claude Opus 5 — flagship Claude for frontier coding, reasoning and agents; 1M context, 128k output. |
Claude Opus 4.8 amanai/claude-opus-4.8 | Anthropic Claude Opus 4.8 — frontier coding and long-horizon agentic work; 1M context. |
Claude Opus 4.7 amanai/claude-opus-4.7 | Anthropic Claude Opus 4.7 — high-end coding and deep reasoning; 1M context. |
Claude Opus 4.6 amanai/claude-opus-4.6 | Anthropic Claude Opus 4.6 — balanced Opus-tier coding and analysis; 1M context. |
Claude Sonnet 5 amanai/claude-sonnet-5 | Anthropic Claude Sonnet 5 — fast, capable coding and everyday reasoning; 1M context. |
Claude Sonnet 4.6 amanai/claude-sonnet-4.6 | Anthropic Claude Sonnet 4.6 — fast all-round coding and chat; 1M context. |
Claude Haiku 4.5 amanai/claude-haiku-4.5 | Anthropic Claude Haiku 4.5 — fastest Claude for lightweight, high-volume tasks; 200k context. |
Grok
| Model | Description |
|---|---|
Grok 4.5 amanai/grok-4.5 | xAI Grok 4.5 — multimodal model with text, image, and file input plus web/X search; 500k context. |
Grok 4.6 amanai/grok-4.6 | xAI Grok 4.6 — agentic coding and visual work with text, image, and file input; 500k context. |
GLM
| Model | Description |
|---|---|
GLM 5.3 amanai/glm-5.3 | Z.AI GLM-5.3 — latest-generation model with a 1M context window and up to 128k output tokens. |
GLM 5.3 Flash amanai/glm-5.3-flash | Z.AI GLM-5.3 Flash — efficient multimodal GLM with a 1M context window and up to 128k output tokens. |
GLM 5.2 amanai/glm-5.2 | Z.AI GLM-5.2 — model for long-horizon tasks with a 1M context window and up to 128k output tokens. |
GLM 5.1 amanai/glm-5.1 | Z.AI GLM-5.1 — flagship for long-horizon autonomous tasks (Opus 4.6-aligned); 200k context, 128k output. |
GLM 5.0 Turbo amanai/glm-5.0-turbo | Z.AI GLM-5-Turbo — deeply optimized for OpenClaw-style agent tasks; 200k context, 128k output. |
GLM 5V Turbo amanai/glm-5v-turbo | Z.AI GLM-5V-Turbo — multimodal coding model with text, image, and video input; 200k context, 128k output. |
Meta
| Model | Description |
|---|---|
Muse Spark 1.1 amanai/muse-spark-1.1 | Meta Muse Spark 1.1 — long-context agentic reasoning; 1,048,576-token context and 1M output verified. Amanai's current route accepts text input only. |
Muse Spark 1.2 amanai/muse-spark-1.2 | Meta Muse Spark 1.2 — long-context agentic reasoning; 1,048,576-token context and 1M output verified. Amanai's current route accepts text input only. |
Qwen
| Model | Description |
|---|---|
Qwen 3.8 Max amanai/qwen3.8-max | Alibaba Qwen3.8 Max — multimodal flagship with text, image, and video input; 1M context. |
Qwen 3.7 Max amanai/qwen3.7-max | Alibaba Qwen3.7 Max — strongest Qwen reasoning with built-in tools; 1M context. |
Qwen 3.7 Plus amanai/qwen3.7-plus | Alibaba Qwen3.7 Plus — balanced multimodal performance and cost for large codebases; 1M context. |
Kimi
| Model | Description |
|---|---|
Kimi K3 amanai/kimi-k3 | Moonshot AI Kimi K3 — multimodal model with text, image, and video input; 1M context for coding and deep reasoning. |
Kimi K2.7 amanai/kimi-k2.7 | Moonshot AI Kimi K2.7 — dedicated coding model, reliable long-context instruction following; 256k context. |
Kimi K2.6 amanai/kimi-k2.6 | Moonshot AI Kimi K2.6 — text and image input with thinking and non-thinking modes; 256k context. |
Kimi K2.5 amanai/kimi-k2.5 | Moonshot AI Kimi K2.5 — open-source SoTA in agents, code and visual understanding; 256k context. |
DeepSeek
| Model | Description |
|---|---|
DeepSeek V4 Pro amanai/deepseek-v4-pro | DeepSeek V4 Pro — flagship reasoning model, thinking + non-thinking modes; 1M context, 384k output. |
DeepSeek V4 Pro 0813 amanai/deepseek-v4-pro-0813 | DeepSeek V4 Pro (snapshot 0813) — pinned Pro release; 1M context, 384k output. |
DeepSeek V4 Flash amanai/deepseek-v4-flash | DeepSeek V4 Flash — fast, cost-efficient V4 with thinking modes; 1M context, 384k output. |
DeepSeek V4.1 Flash amanai/deepseek-v4.1-flash | DeepSeek V4.1 Flash — fast, cost-efficient multimodal reasoning with text and image input; 1M context, 128k output. |
DeepSeek V4 Flash 0731 amanai/deepseek-v4-flash-0731 | DeepSeek V4 Flash (snapshot 0731) — pinned release of V4 Flash; 1M context, 384k output. |
MiniMax
| Model | Description |
|---|---|
MiniMax M3 amanai/minimax-m3 | MiniMax M3 — native multimodal model with text, image, and video input; 1M context via MiniMax Sparse Attention. |
MiniMax M2.7 amanai/minimax-m2.7 | MiniMax M2.7 — efficient coding and agent model; 200k context. |
Hunyuan
| Model | Description |
|---|---|
Hunyuan 3 amanai/hy3 | Tencent Hunyuan 3 — text model for general chat, coding, and reasoning; 1M context, 64k output. |
Hunyuan 4 amanai/hy4 | Tencent Hunyuan 4 — long-context text reasoning model; 1M context, 64k output. |
MOD variants
MOD variants add a predefined instruction profile. Their capabilities, context limits, token prices, and provider routing follow the base model. Select the complete MOD model ID to use the variant. Each variant has its own public ID and usage records.
| MOD model ID | Base model |
|---|---|
amanai/deepseek-v4.1-flash-mod | amanai/deepseek-v4.1-flash |
amanai/glm-5.3-flash-mod | amanai/glm-5.3-flash |
amanai/mimo-v2.6-pro-mod | amanai/mimo-v2.6-pro |
Capabilities
Capabilities are model-specific. Read input_modalities, context_length, max_output, and thinking from the authenticated catalog instead of assuming every model supports the same controls.
| Model | Input | Context | Max output | Reasoning |
|---|---|---|---|---|
GLM 5.3 amanai/glm-5.3 | 1M | 128k | auto, low, medium, high, xhigh, max | |
GLM 5.3 Flash amanai/glm-5.3-flash | 1M | 128k | auto, low, medium, high, xhigh, max | |
GLM 5.2 amanai/glm-5.2 | 1M | 128k | auto, low, medium, high, xhigh, max | |
GLM 5.1 amanai/glm-5.1 | 200k | 128k | auto, low, medium, high | |
GLM 5.0 Turbo amanai/glm-5.0-turbo | 200k | 128k | auto, low, medium, high | |
GLM 5V Turbo amanai/glm-5v-turbo | 200k | 128k | auto, low, medium, high | |
Muse Spark 1.1 amanai/muse-spark-1.1 | 1.05M | 1M | auto, none, low, medium, high, xhigh, max | |
Muse Spark 1.2 amanai/muse-spark-1.2 | 1.05M | 1M | auto, none, low, medium, high, xhigh, max | |
Qwen 3.8 Max amanai/qwen3.8-max | 1M | 64k | auto, none, low, medium, high | |
Qwen 3.7 Max amanai/qwen3.7-max | 1M | 64k | auto, none, low, medium, high | |
Qwen 3.7 Plus amanai/qwen3.7-plus | 1M | 64k | auto, none, low, medium, high | |
Kimi K3 amanai/kimi-k3 | 1M | 64k | auto, low, high, max | |
Kimi K2.7 amanai/kimi-k2.7 | 256k | 32k | auto, low, medium, high | |
Kimi K2.6 amanai/kimi-k2.6 | 256k | 32k | auto, low, medium, high | |
Kimi K2.5 amanai/kimi-k2.5 | 256k | 32k | auto, low, medium, high | |
DeepSeek V4 Pro amanai/deepseek-v4-pro | 1M | 384k | auto, low, medium, high | |
DeepSeek V4 Pro 0813 amanai/deepseek-v4-pro-0813 | 1M | 384k | auto, none, low, medium, high, max | |
DeepSeek V4 Flash amanai/deepseek-v4-flash | 1M | 384k | auto, low, medium, high | |
DeepSeek V4.1 Flash amanai/deepseek-v4.1-flash | 1M | 128k | auto, none, low, medium, high, max | |
DeepSeek V4 Flash 0731 amanai/deepseek-v4-flash-0731 | 1M | 384k | auto, none, low, medium, high, max | |
MiniMax M3 amanai/minimax-m3 | 1M | 48k | auto, low, medium, high | |
MiniMax M2.7 amanai/minimax-m2.7 | 200k | 48k | auto, low, medium, high | |
Hunyuan 3 amanai/hy3 | 1M | 64k | auto, low, medium, high | |
Hunyuan 4 amanai/hy4 | 1M | 64k | auto, low, medium, high | |
Grok 4.5 amanai/grok-4.5 | 500k | 64k | auto, minimal, low, medium, high, xhigh | |
Grok 4.6 amanai/grok-4.6 | 500k | 64k | auto, low, medium, high, xhigh | |
GPT 5.5 amanai/gpt-5.5 | 1.05M | 128k | auto, none, low, medium, high, xhigh | |
GPT 6 Astra amanai/gpt-6-astra | 1.05M | 128k | auto, low, medium, high, xhigh, max | |
GPT 5.6 Sol amanai/gpt-5.6-sol | 1.05M | 128k | auto, none, low, medium, high, xhigh, max | |
GPT 5.6 Terra amanai/gpt-5.6-terra | 1.05M | 128k | auto, none, low, medium, high, xhigh, max | |
GPT 5.6 Luna amanai/gpt-5.6-luna | 1.05M | 128k | auto, none, low, medium, high, xhigh, max | |
Claude Fable 5.1 amanai/claude-fable-5.1 | 1M | 128k | auto, low, medium, high, xhigh, max | |
Claude Fable 5 amanai/claude-fable-5 | 1M | 128k | auto, low, medium, high, xhigh, max | |
Claude Opus 5 amanai/claude-opus-5 | 1M | 128k | auto, low, medium, high, xhigh, max | |
Claude Opus 4.8 amanai/claude-opus-4.8 | 1M | 128k | auto, low, medium, high, xhigh, max | |
Claude Opus 4.7 amanai/claude-opus-4.7 | 1M | 128k | auto, low, medium, high, xhigh, max | |
Claude Opus 4.6 amanai/claude-opus-4.6 | 1M | 128k | auto, none, low, medium, high, xhigh, max | |
Claude Sonnet 5 amanai/claude-sonnet-5 | 1M | 128k | auto, low, medium, high, xhigh, max | |
Claude Sonnet 4.6 amanai/claude-sonnet-4.6 | 1M | 128k | auto, none, low, medium, high, xhigh, max | |
Claude Haiku 4.5 amanai/claude-haiku-4.5 | 200k | 64k | auto, low, medium, high, xhigh, max |
Reasoning controls
Use a level advertised in the selected model's thinking array. auto lets the model choose, while none disables reasoning only where explicitly supported. A rejected or unsupported level should be corrected client-side instead of silently guessed.
Example
curl https://api.amanai.sh/v1/chat/completions \
-H "Authorization: Bearer $AMANAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "amanai/glm-5.3",
"messages": [{"role": "user", "content": "Explain semantic caching briefly."}],
"thinking": "medium",
"stream": true
}'See API reference for common authentication, headers, errors, and discovery. See Billing & pricing for every model's current billing unit and rate.