TokenUs

TokenUs 是 OpenAI 兼容的 AI API 中转服务。上游接入十余家供应商、同型号多渠道互备,在售 39+ 款模型,覆盖 GPT、Gemini、DeepSeek、GLM、Kimi、Qwen、MiniMax、混元等主流系列,支持文本、视觉和图像生成。

37 resources

deepseek-v4-flash

Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work. Call it through the chat completions endpoint with a messages array; it supports tool calling and cached-input billing with a 1M-token context window. Use it for high-volume chat, code assistance, and agent loops where cost matters.

View details

deepseek-v4-flash-0731

Official 0731 upgrade to DeepSeek V4 Flash, combining frontier-class coding, tool use, and agentic performance with fast, economical 1M-context inference. Call it through the chat completions endpoint with a messages array. A good default for production chatbots, coding assistants, and tool-using agents.

View details

deepseek-v4-flash-vision-exp

Experimental vision-enabled variant of DeepSeek V4 Flash. It accepts text and image input and returns text, with tool calling and 1M-token context. Call it through the chat completions endpoint with a messages array. Suited for screenshot understanding, document parsing, and multimodal agents at a low price point.

View details

deepseek-v4-pro

Open MoE flagship with million-token context for coding and long agent runs. It supports tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Use it for repository-scale coding, multi-step agents, and long-document analysis that need the strongest DeepSeek quality.

View details

deepseek-v4-pro-0813

The 0813 checkpoint of DeepSeek V4 Pro, an open MoE flagship with million-token context, tool calling, and strong coding and agentic performance. Call it through the chat completions endpoint with a messages array. Choose this pinned checkpoint when you need reproducible behavior for evaluation or production workloads.

View details

deepseek-v4.1-flash

DeepSeek V4.1 Flash, a fast open-weights model for economical reasoning, coding, and long-context work with a 1M-token window and tool calling. Call it through the chat completions endpoint with a messages array. It is a strong low-cost default for chatbots, RAG pipelines, and batch coding jobs.

View details

deepseek-v4.1-flash-1017

The 1017 checkpoint of DeepSeek V4.1 Flash: fast, economical reasoning and coding with a 1M-token context window and tool calling. Call it through the chat completions endpoint with a messages array. Use this pinned version when you need stable, reproducible outputs for evaluation harnesses or production traffic.

View details

glm-5.1

Flagship GLM model for hybrid reasoning, coding, and agentic engineering with a 200K-token context. It supports tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. A good fit for enterprise assistants, code review, and structured reasoning workloads.

View details

glm-5.2

Open flagship GLM for long-horizon coding agents and million-token context work. It supports tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Use it for large-scale refactoring, autonomous agent loops, and long-document question answering.

View details

glm-5.3

Latest flagship in the GLM 5 series, tuned for reasoning, coding, and agentic tool use with cached-input billing. Call it through the chat completions endpoint with a messages array. It handles complex multi-step instruction following, code generation, and enterprise knowledge assistants with strong accuracy.

View details

glm-5.3-flashx

Extra-fast GLM 5.3 FlashX tier optimized for very low latency and high throughput chat and coding workloads, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Use it for real-time assistants, autocomplete-style coding, and agent fan-out calls.

View details

gpt-5.6-sol

OpenAI GPT-5.6 Sol tier: strong general reasoning, coding, and agentic tool use with vision input and cached-input billing. Call it through the chat completions endpoint with a messages array. A solid pick for production assistants, analysis pipelines, and tool-using agents that need OpenAI-series quality.

View details

gpt-5.6-terra

OpenAI GPT-5.6 Terra tier: a more economical GPT-5.6 variant for chat, coding, and agent workloads, with vision input, tool calling, and cached-input billing. Call it through the chat completions endpoint with a messages array. Use it for high-volume assistants that still need strong quality.

View details

gpt-6-astra

OpenAI GPT-6 Astra, the top GPT-6 tier for the most demanding reasoning, coding, and multimodal agent work, with vision input, tool calling, and cached-input billing. Call it through the chat completions endpoint with a messages array. Choose it when maximum quality justifies the premium price.

View details

gpt-6-luna

OpenAI GPT-6 Luna tier: a lightweight GPT-6 variant for fast, affordable chat and coding assistance with vision input and tool calling. Call it through the chat completions endpoint with a messages array. A good fit for interactive consumer assistants, triage bots, and high-volume agent subtasks.

View details

gpt-6-sol

OpenAI GPT-6 Sol tier with flagship-grade reasoning, coding, and multimodal agent capability, supporting vision input, tool calling, and cached-input billing. Call it through the chat completions endpoint with a messages array. Use it for the hardest production tasks in the GPT-6 family at a moderate price.

View details

gpt-6.1-sol

OpenAI GPT-6.1 Sol tier: an incremental GPT-6 upgrade with stronger reasoning, coding, and multimodal agent performance, plus vision input, tool calling, and cached-input billing. Call it through the chat completions endpoint with a messages array. A strong general-purpose production default.

View details

gpt-6.1-terra

OpenAI GPT-6.1 Terra tier: the economical GPT-6.1 variant balancing price and quality for chat, coding, and agent traffic, with vision input, tool calling, and cached-input billing. Call it through the chat completions endpoint with a messages array. Suited for scaled production deployments.

View details

gpt-image-2

OpenAI GPT Image 2 text-to-image generation. Send a prompt through the images generations endpoint and receive generated images; billed per request. Use it for product mockups, marketing assets, illustrations, and any workflow that needs reliable general-purpose image synthesis from text descriptions.

View details

gpt-image-2.5-flare

OpenAI GPT Image 2.5 Flare, a fast image generation tier for text-to-image tasks, billed per request through the images generations endpoint. Use it for rapid drafts, thumbnails, social assets, and other high-volume image work where turnaround time and cost matter more than maximum fidelity.

View details

gpt-image-2.5-sunburst

OpenAI GPT Image 2.5 Sunburst, a quality-focused image generation tier for text-to-image tasks, billed per request through the images generations endpoint. Use it for final marketing assets, detailed illustrations, and design work that needs the best fidelity in the GPT Image 2.5 family.

View details

hy3

Tencent Hunyuan HY3 open model for general chat, coding, and reasoning at a low price, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. A practical option for Chinese-language assistants, content workflows, and everyday agent tasks.

View details

hy4

Tencent Hunyuan HY4, the newer flagship-class Hunyuan open model for reasoning, coding, and agentic work, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. A strong choice for Chinese-language enterprise assistants and knowledge tasks.

View details

hy4-preview

Preview release of Tencent Hunyuan HY4, the flagship-class open Hunyuan model for reasoning, coding, and agentic workloads, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Use it to evaluate HY4 behavior before the stable release.

View details

kimi-k2.6

Multimodal Kimi workhorse for agent loops, coding tasks, and visual context with a 262K-token window. It supports tool calling, vision input, and cached-input billing. Call it through the chat completions endpoint with a messages array. A balanced choice for production agents that mix text and screenshots.

View details

kimi-k2.7-code

Coding-focused Kimi model, stronger on long-horizon repository work with less overthinking, plus vision input and a 200K-token context. It supports tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Best for coding agents and CI automation.

View details

kimi-k2.8

Kimi K2.8 multimodal model in the K2 series, built for agent loops, coding, and visual understanding with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Use it as a cost-effective upgrade path over earlier K2 releases for production assistants.

View details

kimi-k3

Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work. It supports tool calling, vision input, and cached-input billing. Call it through the chat completions endpoint with a messages array. Suited for deep research, large codebase work, and complex agents.

View details

kimi-k3.1-pro

Kimi K3.1 Pro, the pro tier of the K3 multimodal series for demanding reasoning, coding, and agentic workloads with vision input and tool calling. Call it through the chat completions endpoint with a messages array. Choose it when quality headroom matters more than the price difference over standard tiers.

View details

mimo-v2.5

Xiaomi MiMo V2.5 open-weights model for general chat and coding at a very low price point, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. A practical choice for cost-sensitive assistants, batch jobs, and simple agent loops.

View details

mimo-v2.5-pro

Xiaomi MiMo V2.5 Pro, the stronger pro tier of the MiMo open-weights series for reasoning, coding, and agent tasks, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Use it when the base MiMo tier is not accurate enough.

View details

mimo-v2.6-flash

Xiaomi MiMo V2.6 Flash, the fast low-cost tier of the MiMo V2.6 series for chat, coding, and light reasoning, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Ideal for real-time assistants and high-volume fan-out agent calls.

View details

mimo-v2.6-pro

Xiaomi MiMo V2.6 Pro, the quality tier of the MiMo V2.6 open-weights series for reasoning, coding, and multi-step agent work, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. A strong low-cost alternative to larger flagships.

View details

minimax-m3

MiniMax M3, a large open MoE model for reasoning, coding, and long-context agent workloads, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Well suited for autonomous agents, document analysis, and cost-effective production assistants.

View details

minimax-m3.1-flash

MiniMax M3.1 Flash, the fast economical tier of the M3.1 series for chat, coding, and agent routing, with tool calling and cached-input billing. Call it through the chat completions endpoint with a messages array. Use it for latency-sensitive assistants and high-throughput batch workloads.

View details

qwen3.8-max

Qwen's strongest 2.4T flagship for coding, multimodal agents, professional work, and reliable 1M-context long-horizon tasks. It supports tool calling, vision input, and cached-input billing. Call it through the chat completions endpoint with a messages array. A strong default for enterprise assistants.

View details

qwen3.8-max-0902

The 0902 pinned checkpoint of Qwen 3.8 Max, the 2.4T flagship for coding, multimodal agents, and 1M-context long-horizon tasks, with tool calling and vision input. Call it through the chat completions endpoint with a messages array. Use this dated version for reproducible production behavior.

View details