Moonshot AI API

Kimi K3 is Moonshot AI's flagship open-weight model, released July 16, 2026, with full weights following on July 27. At roughly 2.8 trillion parameters in a Mixture-of-Experts architecture, Moonshot positions it as the largest open-source model released to date, built on two new components: Kimi Delta Attention, a hybrid linear attention mechanism, and Attention Residuals, a replacement for standard residual connections. It runs in an always-on thinking mode with a 1-million-token context window and accepts text, image, and video input. Reported results include 93.5% on GPQA Diamond, 91.2% on BrowseComp, 88.3% on Terminal-Bench 2.1, and a first-place finish on Arena.ai's Frontend Code Arena, putting it close to Claude Opus 4.8 and GPT-5.5 on several agentic and coding tasks. These figures come from Moonshot and early testers, not independently confirmed leaderboards. It suits developers building long-horizon coding agents and tool-calling pipelines who want frontier-level performance at open-weight pricing.

Kimi K2.7 Code Fast is the high-speed serving tier of Kimi K2.7 Code, Moonshot AI's coding-agent model. The weights and capabilities are the same as the standard route; the difference is throughput. Moonshot reports around 180 tokens per second on coding tasks with median-length inputs and up to 260 tokens per second on shorter-context tasks, up to 6x faster than the standard release. Like the base model, it runs in always-on thinking mode with a 262K-token context window and supports tool calling for agentic workflows. Compared with K2.6, K2.7 Code improves coding and agent performance while using roughly 30% fewer thinking tokens. The fast route costs about twice as much per token as the standard endpoint, so it fits interactive coding agents and IDE assistants where response speed matters more than cost.

Kimi K2.7 Code is Moonshot AI's open-weight coding-agent model, released June 2026 and purpose-built for long-horizon, autonomous coding tasks. It shares the same 1-trillion-parameter Mixture-of-Experts architecture (32B active parameters) as K2.6 but is entirely focused on software engineering workloads. Compared to K2.6, it improves 21.8% on Kimi Code Bench v2, 11% on Program Bench, and 31.5% on MLS Bench Lite, while cutting reasoning-token usage by roughly 30%. It always runs in thinking mode — non-thinking mode is not supported. With a 262K-token context window, K2.7 Code is well-suited for multi-file, repository-scale coding pipelines and agentic workflows where sustained reasoning and deep code understanding matter.

Kimi K2.7 Code Highspeed is Moonshot AI's accelerated serving route for Kimi K2.7 Code, its open-weight coding-agent model. It runs the same weights as the standard route; the difference is inference speed, not capability. Moonshot's launch announcement reported throughput of around 180 tokens per second on coding tasks with median-length inputs and up to 260 tokens per second on shorter-context tasks. Access rolled out first to Kimi Code Beta Program members, Kimi API developers, and Kimi Business users, with Moonshot noting capacity limits at launch. Like the base model, it runs in always-on thinking mode with a 262K-token context window, supports tool calling, and accepts text, image, and video input. Input costs $1.90 per million tokens and output $8, making it a fit for interactive coding agents and IDE assistants where response speed matters more than per-token cost.

Kimi K2.6 is Moonshot AI's latest open-weight multimodal model, built on a 1-trillion-parameter mixture-of-experts architecture with a 256K context window. It excels at agentic coding and long-horizon execution, supporting sustained autonomous workflows with 4,000+ tool calls across languages like Rust, Go, and Python. On key benchmarks, it scores 58.6 on SWE-Bench Pro, 54.0 on HLE with Tools, and 50.0 on Toolathlon — competitive with GPT-5.4 and Claude Opus 4.6 on coding and agent tasks, though trailing them on pure reasoning. The model accepts text, image, and video input, supports both thinking and non-thinking modes, and offers an OpenAI-compatible API. It's a strong pick for developers building multi-step agentic workflows and complex software engineering pipelines.

Kimi K2.5 is Moonshot AI's most capable open-source model, a natively multimodal (vision + text) trillion-parameter MoE with 32B active parameters released in January 2026. Built through continual pretraining on ~15 trillion mixed visual and text tokens atop the K2 base, it supports both thinking and instant modes with a 256K context window. It scored 76.8% on SWE-bench Verified, 96.1% on AIME 2025, and 50.2% on Humanity's Last Exam with tools — outperforming Claude Opus 4.5 and GPT-5.2 on the latter. Its standout feature is Agent Swarm, which coordinates up to 100 parallel sub-agents for complex tasks. K2.5 excels at vision-to-code generation, frontend development from screenshots, and large-scale agentic workflows, making it a strong choice for developers building multimodal AI agents.

Kimi K2 Thinking is Moonshot AI's reasoning-enhanced variant of Kimi K2, trained to interleave step-by-step chain-of-thought with dynamic tool calls. It supports up to 200–300 sequential tool calls without drift, enabling deep autonomous research, coding, and analysis workflows. It achieves 71.3% on SWE-bench Verified, 44.9% on Humanity's Last Exam (with tools), 60.2% on BrowseComp, and 99.1% on AIME 2025 (with Python) — placing it among the top open-source thinking models. It uses native INT4 quantization and a 256K context window. K2 Thinking is designed for complex, multi-step tasks where extended reasoning and sustained tool orchestration matter more than low-latency responses.

Kimi K2 0905 is Moonshot AI's September 2025 update to the original Kimi K2, delivering enhanced coding performance and improved tool-calling reliability. It shares the same 1-trillion-parameter MoE architecture with 32B active parameters but doubles the context window from 128K to 256K tokens. Key improvements include stronger frontend development capabilities — producing cleaner, more polished UI code for frameworks like React, Vue, and Angular — along with better integration across popular agent scaffolds. It scored 53.7% Pass@1 on LiveCodeBench. This version is ideal for developers who want K2's agentic strengths with improved real-world coding quality and longer context support for large codebases.

Kimi K2 Turbo Preview is Moonshot AI's high-speed serving tier for Kimi K2, the same trillion-parameter Mixture-of-Experts model as the standard endpoint, run on faster inference infrastructure. Moonshot launched it in August 2025 at 40 tokens per second output, four times the standard endpoint at the time, then raised it later that month to around 60 tokens per second with peaks up to 100. Because the weights are identical, it keeps K2's strengths in agentic tool calling and code generation, with a 256K context window and an OpenAI-compatible API that supports function calling. The tradeoff is price. At $1.15 per million input tokens and $8 per million output tokens, it costs roughly two to three times the standard K2 endpoint. It suits latency-sensitive workloads such as interactive chat, coding assistants, and agent loops where output speed is the bottleneck.

Kimi K2 is a trillion-parameter Mixture-of-Experts model by Moonshot AI, activating 32 billion parameters per token. Designed as a non-thinking model optimized for agentic capabilities, it excels at tool use, code generation, and autonomous problem-solving with a 128K token context window. On benchmarks, K2 scored 65.8% on SWE-bench Verified, 75.1% on GPQA-Diamond, 49.5% on AIME 2025, and 66.1 on Tau2-bench — surpassing most open- and closed-source models in non-thinking settings. It ranked as the #1 open-source model on the LMSYS Arena leaderboard upon release in July 2025. K2 is well suited for developers building AI agents and tool-calling pipelines who need strong coding and reasoning without extended thinking overhead.

Kimi K2 0711 Preview is the July 11, 2025 snapshot of Moonshot AI's Kimi K2, a non-thinking Mixture-of-Experts model built for agentic tasks and served through Moonshot's OpenAI-compatible API. It targets tool use, code generation, and multi-step autonomous workflows within a 128K token context window. On benchmarks, it scored 65.8% on SWE-bench Verified, 66.1 on Tau2-bench, 76.5 on ACEBench (en), 75.1% on GPQA-Diamond, and 49.5% on AIME 2025. Moonshot's technical report shows it outperforming most open- and closed-weight models in non-thinking settings, and it ranked as the top open-source model on the LMArena leaderboard at release. As a dated snapshot, its behavior stays fixed while later updates (such as the 0905 version) ship separately, which suits developers who want reproducible results from the original K2 in agent and tool-calling pipelines.

Kimi K2 Instruct is the instruction-tuned version of Moonshot AI's Kimi K2, released in July 2025 as a general-purpose chat and agentic model. Moonshot describes it as reflex-grade, meaning it answers directly without extended thinking, which keeps latency low in tool-calling loops. It scores 65.8% on SWE-bench Verified with agentic tools (single attempt), 70.6% on Tau2-bench retail, and 76.5% on AceBench, close to Claude 4 on tool-use benchmarks at the time of release. Tool calling is built in; you pass a list of available tools and the model decides when and how to invoke them. This route offers a 128K context window. It suits developers building agents, coding assistants, and multi-step tool pipelines that need autonomous execution without chain-of-thought token overhead.

Kimi Dev 72B is a 72-billion-parameter coding model by Moonshot AI, purpose-built for software engineering tasks like bug fixing, code generation, and unit test creation. It is based on the Qwen 2.5-72B architecture and fine-tuned with large-scale reinforcement learning on real-world GitHub issues and pull requests. The model achieved 60.4% on SWE-bench Verified, setting a state-of-the-art result among open-source models at the time of its June 2025 release. It uses a two-stage framework — file localization followed by precise code editing — that mirrors how human developers approach issue resolution. Kimi Dev 72B is a strong pick for automated code repair and test generation workflows where a specialized coding model outperforms general-purpose alternatives.

Moonshot V1 8K Vision Preview is a multimodal variant of Moonshot AI's V1 model that accepts both image and text inputs within an 8,000-token context window. It can interpret screenshots, charts, UI mockups, and photos, returning text-based analysis. This makes it useful for tasks like image captioning, visual Q&A, and lightweight document understanding where the source material includes visual elements. As a preview model, it may see changes before a stable release. The API follows the OpenAI-compatible content array format with image_url blocks, making integration straightforward for developers already using similar patterns.

Moonshot V1 32K Vision Preview is a multimodal model from Moonshot AI that processes both images and text within a 32,000-token context window. It extends the base 32K model with the ability to interpret visual inputs — including screenshots, diagrams, charts, and scanned documents — and return text-based responses. This is useful for workflows that combine visual context with moderate-length text, such as analyzing annotated documents or explaining UI designs. As a preview release, the vision capabilities may evolve. The API accepts the standard OpenAI-compatible content array format for multimodal inputs.

Moonshot V1 128K Vision Preview is Moonshot AI's largest-context multimodal model in the V1 series, supporting both image and text inputs within a 128,000-token context window. It combines the long-context strength of the 128K text model with visual understanding capabilities. This makes it well-suited for processing large multimodal documents — think lengthy reports with embedded charts, multi-page scanned PDFs, or extensive UI review sessions. As a preview model, vision features may be refined over time. The API uses the standard OpenAI-compatible format for multimodal content, making it a drop-in addition to existing workflows.

Kimi V1 128K Vision Preview is a multimodal chat model from Moonshot AI that accepts image and text input and returns text. It is the kimi-v1 listing of the model previously published as moonshot-v1-128k-vision-preview, matching that route's context window and pricing, and reflects the move of Moonshot's developer platform to the Kimi brand. The model reads visual content in images, including embedded text, colors, and object shapes, which makes it suitable for image description, visual question answering, and document understanding. The 128K context tier is the largest in the V1 vision line, sized for long multimodal inputs such as multi-page scanned documents. The API is OpenAI-compatible, taking images through the standard content array with image_url blocks, and supports streaming, tool invocation, and JSON mode. It remains a preview release, and Moonshot's newer K-series models are now the company's main line.

Kimi V1 32K Vision Preview is a multimodal chat model from Moonshot AI, the Beijing company behind the Kimi assistant. The kimi-v1 model ID refers to the same V1 line that Moonshot's Kimi platform also lists as moonshot-v1; this route serves the 32K vision preview tier of that family. The vision preview accepts image input alongside text and can read text in images, identify colors, and describe object shapes, which covers screenshot interpretation, chart reading, and lightweight document OCR. Images are billed as tokens based on resolution, and Moonshot notes that resolutions above 4K add cost without improving understanding. The 32K context tier sits between the 8K and 128K variants, fitting multi-turn conversations and medium-length documents that mix visuals with text. The API is OpenAI-compatible and supports streaming and tool calling.

Kimi V1 8K Vision Preview is a multimodal chat model from Moonshot AI, the company behind the Kimi assistant, accepting image and text input within an 8,192-token context window. This route uses the kimi-v1 name for the model line Moonshot's platform lists as moonshot-v1-8k-vision-preview; the context window and pricing match. Moonshot's vision models recognize text in images (OCR), colors, and object shapes, which covers screenshots, charts, and scanned documents. Images are passed as base64-encoded image_url blocks in the OpenAI-compatible content array format; URL-referenced images are not supported. Multi-turn conversation, streaming, and tool calling are all supported. The 8K tier is the smallest, lowest-cost context size in the V1 vision line, suited to single images with short prompts, such as captioning, visual Q&A, and light document extraction. As a preview route, behavior may change before a stable release.

Moonshot V1 Auto is a smart routing layer from Moonshot AI that automatically selects the most cost-efficient context window — 8K, 32K, or 128K — based on the token count of each request. It uses the same underlying Moonshot V1 model as the fixed-context variants, so there is no difference in output quality. The routing simply ensures you're billed at the lowest applicable tier for each call, eliminating the need to manually choose a context size or overpay for unused capacity. Usage is identical to the other Moonshot V1 models — just set the model ID to `moonshot-v1-auto` and the platform handles the rest. Ideal for applications with variable-length inputs.

Moonshot V1 8K is a general-purpose text generation model from Moonshot AI, the Beijing-based company behind the Kimi assistant. It supports an 8,000-token context window, making it the most lightweight option in the Moonshot V1 family. All Moonshot V1 models share the same underlying capabilities — the only difference is the maximum context length. This variant is best suited for short-form tasks like single-turn Q&A, classification, and concise summaries where you want to minimize token costs. The API is OpenAI-compatible, so you can integrate it by swapping the base URL and API key in any existing OpenAI SDK setup. The model handles both English and Chinese well.

Moonshot V1 32K is a general-purpose text generation model from Moonshot AI with a 32,000-token context window. It sits in the middle of the Moonshot V1 family, balancing context capacity with cost. All Moonshot V1 variants share the same model quality — only the context length differs. The 32K window is well-suited for multi-turn conversations, medium-length document summarization, and tasks where inputs and outputs together exceed 8K tokens but don't require the full 128K capacity. The API is fully OpenAI-compatible, supporting streaming, tool calling, and standard chat completion parameters. The model performs well in both English and Chinese.

Moonshot V1 128K is a long-context text generation model from Moonshot AI, offering a 128,000-token context window. Moonshot AI was one of the first companies to ship native 128K-token context support when the Kimi chatbot launched in 2023. This variant is designed for tasks that demand large input windows: processing entire codebases, analyzing lengthy legal or financial documents, or maintaining very long conversation histories. It shares the same model quality as the 8K and 32K variants — context length is the only differentiator. The API is OpenAI-compatible and supports streaming, tool calling, and context caching for reduced latency and cost on repeated prompts.

Kimi V1 128K is the 128,000-token context tier of Moonshot AI's original v1 chat model line, the same route previously listed as moonshot-v1-128k before Moonshot rebranded its developer platform around the Kimi name. The 8K, 32K, and 128K tiers share the same underlying model, and context length is the only difference between them. Moonshot's docs describe the 128K tier as suited to generating very long texts, which makes it a fit for long documents, large transcripts, and extended conversation histories. The API is OpenAI-compatible and supports tool calling. This is a legacy line. Moonshot's newer K-series models (Kimi K2 and K3) now cover agentic and reasoning workloads, and the v1 series is no longer offered to newly registered platform accounts, so this route mainly serves existing integrations.

Kimi V1 32K is a general-purpose chat model from Moonshot AI, the company behind the Kimi assistant, served under the kimi-v1 name. It corresponds to the moonshot-v1-32k model on Moonshot's platform, where the V1 tiers (8K, 32K, 128K) share the same underlying model and differ only in maximum context length. The 32K tier provides a 32,768-token context window with matching max output. It fits multi-turn conversations and medium-length documents that outgrow the 8K tier without needing the 128K window. The API is OpenAI-compatible, supports tool calling, and the model handles both English and Chinese. Moonshot has since moved its platform to the Kimi K2 and K3 series, and the V1 line is no longer open to newly registered users there. It remains an option for existing workloads that already target this model.

Kimi V1 8K is a general-purpose chat model from Moonshot AI, the Chinese company behind the Kimi assistant. The kimi-v1 name follows the Kimi branding Moonshot now uses for its API platform, and the route matches the original moonshot-v1 8K tier, with the same 8,192-token context window and pricing. Like the rest of the v1 family, the variants differ only by context length. The 8K tier is the smallest and cheapest, aimed at short tasks such as single-turn Q&A, classification, and brief summaries where inputs and outputs fit in a small window. The API is OpenAI-compatible, so it works with existing OpenAI SDK setups by changing the base URL and API key. Input costs $0.20 per million tokens and output $2, which suits high-volume, short-context workloads.