Kimi K2.7 Code: Complete Specifications, Pricing, API Access & Use Cases (2026)
What Is Kimi K2.7 Code?
Kimi K2.7 Code is Moonshot AI’s open-source, coding-focused agentic model, released on the official Kimi resource page dated June 25, 2026, featuring a 262,144-token context window and long-horizon coding capability, with Kimi API pricing of \$0.95 per 1M cache-miss input tokens and \$4.00 per 1M output tokens as of July 2026.
Moonshot AI describes Kimi K2.7 Code as a model built for complex software engineering workflows, repository-scale context, coding agents, and instruction-following across long sessions. It belongs to the Kimi K2 family and is built on a Mixture-of-Experts architecture with 1T total parameters and 32B activated parameters per token.
Users typically search for Kimi K2.7 Code when they want model details for agentic coding, long-context programming, multimodal code-related workflows, or API cost planning. It is related to the broader Kimi research line, including Kimi k1.5 model research, but K2.7 Code is a later coding-focused API and open-weight model.
What Are Kimi K2.7 Code’s Key Specifications and Pricing?
| Field | Verified Value |
|---|---|
| Provider | Moonshot AI / Kimi (as of July 2026). |
| Model Family | Kimi K2 family (as of July 2026). |
| Model Type | Open-source, coding-focused, agentic, multimodal MoE reasoning model (as of July 2026). |
| Release Date | June 25, 2026 official Kimi resource page date (as of July 2026). |
| Context Window | 262,144 tokens / 256K context length (as of July 2026). |
| Input Pricing | \$0.95 per 1M cache-miss input tokens on Kimi API (as of July 2026). |
| Cached Input Pricing | \$0.19 per 1M cache-hit input tokens on Kimi API (as of July 2026). |
| Cache Write Pricing | \$1.11 per 1M tokens in the Gate.AI model data; not separately confirmed in the checked Kimi official pricing page (as of July 2026). |
| Output Pricing | \$4.00 per 1M output tokens on Kimi API (as of July 2026). |
| Pricing Unit | 1M tokens, where 1M means 1,000,000 tokens (as of July 2026). |
| Modality Support | Text, image, and video input; text output verified through chat-completion examples (as of July 2026). |
| Supported Input Types | Text, image, video (as of July 2026). |
| Supported Output Types | Text response output; image, audio, or video generation output not confirmed from official sources as of July 2026. |
| API Access | Kimi API, Kimi Code, Hugging Face weights, and Gate.AI access using model ID (as of July 2026). |
| Provider API Model ID | kimi-k2.7-code (as of July 2026). |
| Gate.AIModel ID | moonshotai/kimi-k2.7-code, from Gate.AI model data (as of July 2026). |
| Availability | Kimi Code, Kimi API, Hugging Face model weights, and Gate.AI via model data (as of July 2026). |
| Knowledge Cutoff | Not confirmed from official sources as of July 2026. |
| Rate Limits | Not confirmed for this specific model from official sources as of July 2026. |
| Fine-tuning Support | Not confirmed from official sources as of July 2026. |
| Streaming Support | Kimi API uses OpenAI-compatible Chat Completions; model-specific streaming limits not confirmed as of July 2026. |
| Batch API Support | Kimi API documents Batch API endpoints, but model-specific batch availability was not confirmed as of July 2026. |
| Tool / Function Calling | ToolCalls and multi-step tool-call design are documented for K2.7 Code (as of July 2026). |
| Structured Output / JSON Mode | JSON Mode and Partial Mode are listed in Kimi API pricing documentation (as of July 2026). |
| License / Usage Restrictions | Modified MIT License for code repository and model weights (as of July 2026). |
What Can Kimi K2.7 Code Do That Makes It Useful in Production?
Kimi K2.7 Code is most relevant for production teams that need coding agents, long-context code review, and multi-file software changes. Its 262,144-token context window helps with larger repositories, although a large context window does not remove the need for testing, review, and prompt discipline.
First, it can support long-horizon coding workflows such as refactoring, debugging, and feature implementation across multiple files. Moonshot describes K2.7 Code as optimized for long-context coding and improved end-to-end task completion across complex software engineering workflows.
Second, it supports multimodal software tasks. Because K2.7 Code accepts text, image, and video inputs, it may fit workflows involving screenshots, interface recordings, diagrams, or visual bug reports, while still producing text responses.
Third, K2.7 Code uses always-on thinking mode. Official Kimi material states that non-thinking mode is not supported for this model and that thinking is enabled by default, which is relevant for tasks that benefit from deeper reasoning but may increase token usage.
Fourth, it can support agent frameworks and tool workflows. Kimi’s official documentation lists ToolCalls, JSON Mode, Partial Mode, and interleaved thinking with multi-step tool call design, which makes the model relevant to developer agents and structured automation.
What Are Kimi K2.7 Code’s Supported Modalities?
| Modality | Supported? | Notes |
|---|---|---|
| Text input | Yes | Supported through chat-completion usage and coding workflows. |
| Image input | Yes | Official examples show image input through image_url. |
| Video input | Yes | Official examples show video input through video_url; third-party deployments may have limitations. |
| Audio input | Not confirmed | No official K2.7 Code audio-input support was confirmed. |
| Text output | Yes | Chat-completion examples return response content as text. |
| Image output | Not confirmed | K2.7 Code is not documented as an image-generation model. |
| Audio/video output | Not confirmed | No audio or video generation output was verified. |
Where Does Kimi K2.7 Code Fall Short?
Kimi K2.7 Code’s main limitation is that some product details remain incomplete or platform-specific. Knowledge cutoff, model-specific rate limits, fine-tuning support, regional availability, and model-specific batch support were not confirmed from official sources as of July 2026.
The model also has a cost and latency trade-off. Always-on thinking may improve reasoning for complex coding tasks, but it can consume additional tokens. Moonshot says K2.7 Code reduces thinking-token usage compared with K2.6, but users should still budget for long prompts, long context, and multi-step agent workflows.
This is a general AI limitation and is not model-specific unless stated by the provider: Kimi K2.7 Code can still produce incorrect code, insecure suggestions, broken dependencies, or plausible but false explanations. Production use should include automated tests, dependency scanning, human review, and separate checks for legal, medical, financial, or security-sensitive code paths.
Finally, official benchmark tables are provider-reported. They are useful for understanding intended evaluation areas, but they should not be treated as guarantees for a specific codebase, programming language, latency target, or deployment environment.
What Is Kimi K2.7 Code Best Used For?
| Use Case | Why Kimi K2.7 Code May Fit | Important Limitation |
|---|---|---|
| Long-context code review | 262,144-token context can help inspect larger files, logs, or repository excerpts. | Full repositories may still exceed practical context and cost limits. |
| Coding agents | Always-on thinking, ToolCalls, Partial Mode, and multi-step tool-call design are relevant to agent workflows. | Agent output should be sandboxed and tested. |
| UI bug analysis | Image and video inputs can support screenshot- or recording-based debugging. | Output is text; it does not directly generate UI assets. |
| Refactoring and migration planning | Long-context instruction following may help with multi-file changes. | Human review is needed before production merges. |
| Coding-cost evaluation | Kimi API pricing and cache discounts are clearly published. | Long reasoning and long prompts can still increase total cost. |
For teams comparing reasoning and coding models, reasoning model trade-offs and coding-capable multimodal model profiles can provide useful context, but Kimi K2.7 Code should be evaluated on the team’s own codebase before production adoption.
How Does Kimi K2.7 Code Compare to Kimi K2.6 and Claude Opus 4.8?
| Comparison Area | Kimi K2.7 Code | Kimi K2.6 | Claude Opus 4.8 | Scenario Fit |
|---|---|---|---|---|
| Positioning | Coding-focused agentic model built on K2.6. | More general Kimi K2 family model. | External coding-agent comparison point in Moonshot’s benchmark table. | K2.7 Code is more scenario-specific for coding. |
| Context | 262,144 tokens. | 262,144 tokens in Kimi pricing pages. | Not independently verified in this article. | Long-context workflows require direct testing. |
| Pricing | \$0.95 cache-miss input, \$0.19 cache-hit input, \$4.00 output per 1M tokens on Kimi API. | Kimi platform lists \$0.95 input and \$4.00 output for K2.6, with different cache-hit price. | Not verified from Anthropic official sources in this article. | Use verified provider pricing before procurement. |
| Coding benchmarks | Moonshot reports gains over K2.6 on Kimi Code Bench v2, Program Bench, and MLS Bench Lite. | Used as Moonshot’s prior-family baseline. | Included in Moonshot’s benchmark table, but third-party conditions differ. | Benchmarks are directional, not deployment guarantees. |
| Modalities | Text, image, and video input; text output. | Kimi docs describe multimodal input for K2.6. | Not independently verified in this article. | K2.7 Code may fit multimodal bug-report workflows. |
Moonshot’s own benchmark table compares Kimi K2.7 Code with Kimi K2.6, GPT-5.5, and Claude Opus 4.8, but this article treats those results as provider-reported evaluation data rather than an overall model ranking.
How Do I Access Kimi K2.7 Code Through Gate.AI?
Gate.AI documentation verifies an OpenAI-compatible gateway with the base URL https://api.gate.ai/openai/v1, API-key authentication, and Chat Completions support. The Gate.AI model ID and prices for this page come from the Gate.AI model data: moonshotai/kimi-k2.7-code, \$0.95 input, \$4.00 output, \$0.19 cache read, and \$1.11 cache write per 1M tokens as of July 2026.
Gate.AI pricing documentation also describes pay-as-you-go billing, prompt-cache billing when a model supports caching, API key management, smart routing, usage insights, and organization permissions.
Python Example
from openai import OpenAIimport osclient = OpenAI(api_key=os.environ["GATEAI_API_KEY"],base_url="https://api.gate.ai/openai/v1",)completion = client.chat.completions.create(model="moonshotai/kimi-k2.7-code",messages=[{"role": "user","content": "Review this Python function and suggest safer error handling."}],)print(completion.choices[0].message.content)
curl Example
curl https://api.gate.ai/openai/v1/chat/completions \-H "Authorization: Bearer $GATEAI_API_KEY" \-H "Content-Type: application/json" \-d '{"model": "moonshotai/kimi-k2.7-code","messages": [{"role": "user","content": "Write a concise test plan for a repository migration."}]}'
Through Gate.AI, developers can use an OpenAI-compatible access pattern while managing API keys, usage, pricing, prompt-cache visibility, and team permissions where those features are enabled for the account.
FAQs
What is Kimi K2.7 Code’s context window?
Kimi K2.7 Code supports a 262,144-token context window, also described as 256K context length, as of July 2026.
How much does Kimi K2.7 Code cost?
Kimi API pricing is \$0.95 per 1M cache-miss input tokens, \$0.19 per 1M cache-hit input tokens, and \$4.00 per 1M output tokens as of July 2026.
How can developers access Kimi K2.7 Code?
Developers can access it through Kimi API, Kimi Code, Hugging Face weights, and Gate.AI using the Gate.AI model ID moonshotai/kimi-k2.7-code.
What is Kimi K2.7 Code mainly used for?
It is suited for long-horizon coding, repository analysis, coding agents, multimodal bug-report interpretation, refactoring support, and structured developer workflows that require long context and reasoning.
