> ## Documentation Index
> Fetch the complete documentation index at: https://docs.deepshi.ai/llms.txt
> Use this file to discover all available pages before exploring further.
> ## Agent Instructions
> The Deepshi API is an OpenAI-compatible gateway. Base URL: https://api.deepshi.ai/v1. API keys start with sk-bf- and go in the Authorization: Bearer header. Prefer the official OpenAI SDKs pointed at the Deepshi base URL. Model ids are clean with no provider prefix (e.g. deepshi-3.0, claude-opus-4.8, gpt-5.5). Chat and image requests are synchronous; video and music requests are asynchronous job APIs (create, then poll). Every synchronous response carries usage.cost.total_cost in USD. Do not reference the internal /api/* admin plane or virtual keys.
# Chat models
> Every chat model available through the Deepshi API, with context windows and capabilities.
Use these models with the [chat completions](/capabilities/text-and-chat) endpoint. Reference a model by its **id** in the `model` field.
The exact set your key can call is returned by `GET /v1/models`. For per-model
rates, see [deepshi.ai](https://deepshi.ai/).
## Deepshi models
All of Deepshi's own models are uncensored.
`deepshi-2.0` · **Context** 128K
Highly uncensored, with open, minimally-filtered responses. The most direct option.
Tools Reasoning Vision
`deepshi-3.0` · **Context** 256K
Multimodal flagship with strong visual understanding and agentic coding.
Tools Reasoning Vision
## Third-party models
The latest models from leading providers, through the same endpoint, key, and balance.
Recently added models appear first.
`inkling` · **Context** 1M
Thinking Machines' open-weight MoE model for reasoning, coding, and agentic work, with image and audio input.
Tools Reasoning Vision
`kimi-k3` · **Context** 1M
Moonshot's open-weight reasoning model for large codebases, tool use, and long autonomous runs.
Tools Reasoning Vision
`claude-sonnet-5` · **Context** 1M
Anthropic's most capable Sonnet, balancing speed and depth for coding and agent workflows.
Tools Reasoning Vision
`claude-fable-5` · **Context** 1M
Anthropic's agentic Claude 5 model for long-running, minimally-supervised coding and research.
Tools Reasoning Vision
`gpt-5.6-sol` · **Context** 1M
The flagship of OpenAI's GPT-5.6 series, built for complex reasoning and multi-step coding.
Tools Reasoning Vision
`gpt-5.6-terra` · **Context** 1M
The balanced GPT-5.6 tier for everyday coding, reasoning, and agentic tasks.
Tools Reasoning Vision
`gpt-5.6-luna` · **Context** 1M
The fast, cost-efficient GPT-5.6 tier for high-volume, latency-sensitive work.
Tools Reasoning Vision
`grok-4.5` · **Context** 500K
xAI's most capable model, with frontier performance across coding, knowledge work, and STEM.
Tools Reasoning Vision
`kimi-k2.7-code` · **Context** 256K
Moonshot's coding-focused Kimi model for end-to-end programming over long contexts.
Tools Reasoning Vision
`nemotron-3-ultra-550b-a55b` · **Context** 1M
NVIDIA's open frontier reasoning and orchestration model with a 1M context.
Tools Reasoning
`minimax-m3` · **Context** 1M
MiniMax multimodal model for long-horizon agentic work, with a 1M context.
Tools Reasoning Vision
`step-3.7-flash` · **Context** 256K
StepFun's efficient multimodal model with native image and video understanding.
Tools Reasoning Vision
`grok-build-0.1` · **Context** 256K
xAI's fast coding model built for agentic software engineering.
Tools Reasoning Vision
`granite-4.1-8b` · **Context** 128K
IBM's dense 8B model for enterprise retrieval, tools, and structured text.
Tools
`qwen3.6-35b-a3b` · **Context** 256K
Alibaba's open multimodal MoE balancing quality and efficient inference.
Tools Reasoning Vision
`qwen3.6-27b` · **Context** 256K
Dense 27B Qwen model with multimodal input for general-purpose work.
Tools Reasoning Vision
`deepseek-v4-pro` · **Context** 1M
DeepSeek's large MoE for advanced reasoning and coding over a 1M context.
Tools Reasoning
`deepseek-v4-flash` · **Context** 1M
Efficiency-tuned DeepSeek MoE for fast, low-cost reasoning at a 1M context.
Tools Reasoning
`mimo-v2.5-pro` · **Context** 1M
Xiaomi's flagship for agentic capability and complex software engineering.
Tools Reasoning
`gpt-5.4-mini` · **Context** 400K
Faster, efficient GPT-5.4 variant for high-throughput workloads.
Tools Reasoning Vision
`gpt-5.4-nano` · **Context** 400K
The lightest GPT-5.4 model, tuned for low-latency, high-volume use.
Tools Reasoning Vision
`glm-5-turbo` · **Context** 256K
Z.ai's fast-inference model tuned for agent-driven workflows.
Tools Reasoning
`nemotron-3-super-120b-a12b` · **Context** 1M
NVIDIA's 120B hybrid MoE for efficient multi-agent applications.
Tools Reasoning
`minimax-m2.5` · **Context** 200K
MiniMax model trained for real-world productivity and coding.
Tools Reasoning
`qwen3-coder-next` · **Context** 256K
Qwen's open coding model for coding agents and local development.
Tools
`glm-4.7-flash` · **Context** 200K
Z.ai's 30B-class model balancing performance with agentic coding.
Tools Reasoning
`gemma-4-26b-a4b-it` · **Context** 256K
Google's instruction-tuned MoE delivering near-31B quality at lower cost.
Tools Reasoning Vision
`minimax-m2.7` · **Context** 200K
Next-generation MiniMax model for autonomous, agentic productivity.
Tools Reasoning
`gpt-oss-120b` · **Context** 128K
OpenAI's open-weight 117B MoE for high-reasoning, agentic use.
Tools Reasoning
`gpt-oss-20b` · **Context** 128K
OpenAI's open-weight 21B MoE under Apache 2.0, tuned for efficient reasoning.
Tools Reasoning
`claude-opus-4.8` · **Context** 1M
Anthropic's most capable Opus model, with deep reasoning over long, complex tasks.
Tools Reasoning Vision
`claude-opus-4.7` · **Context** 1M
Opus-family model built for long-running, asynchronous agents.
Tools Reasoning Vision
`claude-sonnet-4.6` · **Context** 1M
Strong all-rounder for coding, agents, and professional work.
Tools Reasoning Vision
`gpt-5.5` · **Context** 1M
OpenAI's frontier model for complex professional workloads.
Tools Reasoning Vision
`gpt-5.4` · **Context** 1M
Frontier model unifying the GPT and Codex lines, strong at agentic coding.
Tools Reasoning Vision
`gpt-4.1` · **Context** 1M
Tuned for precise instruction-following and software engineering.
Tools Vision
`gpt-4o` · **Context** 128K
GPT-4-class intelligence that runs faster and cheaper. Solid general-purpose pick.
Tools Vision
`gemini-3.5-flash` · **Context** 1M
Google's high-efficiency model with near-Pro reasoning at Flash speed and cost.
Tools Reasoning Vision
`grok-4.3` · **Context** 1M
xAI reasoning model suited to agentic workflows and high factual accuracy.
Tools Reasoning Vision
`grok-4.20` · **Context** 2M
Fast xAI reasoning model with strong tool-calling and low hallucination.
Tools Reasoning Vision
`glm-5.2` · **Context** 1M
Large-scale reasoning model for long-horizon agent and engineering work.
Tools Reasoning
`glm-5.1` · **Context** 200K
A major step up in coding ability on long-horizon tasks.
Tools Reasoning
`glm-5` · **Context** 200K
Z.ai's flagship open model for complex systems and agent workflows.
Tools Reasoning
`kimi-k2.6` · **Context** 256K
Built for long-horizon coding, UI/UX generation, and multi-agent orchestration.
Tools Reasoning Vision
`gemma-4-31b-it` · **Context** 256K
A \~31B dense open model with an optional reasoning mode and native tools.
Tools Reasoning Vision
Not sure where to start? Use **`deepshi-2.0`** for highly uncensored
responses, or **`deepshi-3.0`** for an all-round flagship with tools and
reasoning.