Kimi K2.7 Code
Open-weight, coding-focused agentic model with 1T parameters, 256K context, and strong performance on long-horizon software tasks.
Overview
What is Kimi K2.7 Code
Kimi K2.7 Code is Moonshot AI's open-weight, coding-focused agentic model released on June 12, 2026. Built on Kimi K2.6, it improves long-horizon coding performance, reduces reasoning-token usage by ~30%, and supports vision, function calling, web search, and code optimization within a 256K-token context.
Running it privately on Venice
On Venice, Kimi K2.7 Code runs with zero retention — your prompts are never stored or profiled. This open, uncensored model executes in a private environment, preserving your sovereignty while enabling tool use, vision, and web-connected reasoning. You maintain full control over sensitive code and agent workflows.
Assessment
Strengths and limitations
- Open-weight and commercially usable under Modified MIT license, with no restrictions below 100M users or $20M monthly revenue.
- Highly efficient reasoning: reduces thinking-token usage by ~30% compared to K2.6, lowering cost per task.
- Strong performance on long-horizon coding benchmarks: +21.8% on Kimi Code Bench v2, +11.0% on Program Bench.
- Integrated vision, web search, and function calling enable autonomous agent workflows.
- Supports 256K context for complex, multi-file software engineering tasks.
- 256K context is smaller than rivals like Kimi K3 or Claude models offering 1M tokens.
- Benchmarks are self-reported; independent verification is limited.
- Cannot disable thinking mode: full chain-of-thought runs on every request, limiting control over output style.
- Vision and web capabilities are strong but not as refined as top-tier proprietary models.
Capabilities
What it supports
- Tool use / function calling
- Vision (image input)
- Reasoning
- Web search
- Code-optimized
- Structured output (JSON schema)
- Audio input
- Video input
- Multiple image inputs
- Log probabilities
Specifications
Datasheet
- Maker
- Moonshot AI
- Released
- June 12, 2026
- Architecture
- Mixture-of-Experts (MoE)
- Total parameters
- 1T
- Activated parameters
- 32B
- Open weights
- Yes — Modified MIT license
- Context window
- 256K tokens
- Max output
- 65.536K tokens
- Capabilities
- Vision, Function calling, Reasoning, Web search, Code-optimized
- Privacy on Venice
- Private — zero retention
- Available on Venice since
- Jun 2026
- License
- Modified MIT
API
Call it from your code
Venice exposes an OpenAI-compatible API. Point your base URL at Venice and pass the model id.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k2-7-code",
"messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
}'Pricing
What it costs on Venice
Billed per token on Venice: $0.75 per 1M input tokens and $3.50 per 1M output tokens.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
Alternatives
How it compares
| Model | Max resolution | Strongest at | Open weights | Price (Venice) |
|---|---|---|---|---|
| Kimi K2.7 Code | N/A | Long-horizon coding | Yes | $0.75 in · $3.50 out / 1M |
| Kimi K3 | N/A | Ultra-long context (1M) | Yes | $3.75 in · $18.75 out / 1M |
| GLM 5.1 | N/A | Balanced coding & reasoning | Yes | $1.10 in · $4.15 out / 1M |
| Claude Sonnet 4.6 | N/A | General reasoning | No | $3.60 in · $18 out / 1M |
Open-weight, efficient coding agent with strong agentic gains over K2.6.
Use cases
What it is good for
- 01End-to-end software development: planning, coding, debugging, and refactoring across large codebases.
- 02Autonomous coding agents that use tools, search documentation, and execute commands.
- 03Code review and optimization with long-context awareness across multiple files.
- 04Vision-enhanced programming tasks, such as interpreting screenshots of UIs or diagrams.
- 05Private, on-prem or sovereign deployments using open weights and Venice’s zero-retention guarantee.
Prompting
Getting better results
Use explicit, step-by-step instructions — the model excels at following structured plans.
Include file paths and context when editing multi-file projects to maintain coherence.
Leverage web search and function calling for up-to-date library documentation or API references.
For vision tasks, describe image content in detail even if the image is attached — improves robustness.
Use the high-speed variant when latency matters, but expect slight fluctuations in availability.
Version history
Predecessor model.
Current — optimized for coding and agent tasks.
FAQ
Frequently asked questions
Kimi K2.7 Code is Moonshot AI's open-weight, coding-focused agentic model released on June 12, 2026. It improves long-horizon software engineering performance over K2.6, reduces reasoning-token usage by ~30%, and supports vision, web search, and function calling within a 256K context.
Yes. Kimi K2.7 Code is open-weight under a Modified MIT license, meaning the model weights are freely available for use, modification, and redistribution. Commercial use is allowed, with a requirement to credit 'Kimi K2' if your product exceeds 100M monthly active users or $20M in monthly revenue.
On Venice, Kimi K2.7 Code costs $0.75 per 1M input tokens and $3.50 per 1M output tokens. Cached input is even cheaper at $0.16 per 1M tokens, making repeated access to long contexts cost-efficient.
Yes. Kimi K2.7 Code supports multiple image inputs using the MoonViT vision encoder, enabling it to interpret screenshots, diagrams, and UIs as part of coding or agent workflows.
Yes. Kimi K2.7 Code supports tool use and function calling, allowing it to execute code, search the web, and interact with external systems as part of autonomous agent workflows.
Kimi K2.7 Code supports a context window of 256,000 tokens, suitable for large codebases and long-horizon tasks, though smaller than some rivals offering 1M tokens.
Kimi K3 offers a 1M-token context and stronger overall performance but at 5x the cost. Kimi K2.7 Code is more efficient and open, making it better for budget-conscious developers needing strong coding performance without the largest context.
Yes. On Venice, Kimi K2.7 Code runs with zero retention — your prompts are never stored, profiled, or used for training. This ensures full privacy and sovereignty, ideal for sensitive or proprietary development workflows.
Run Kimi K2.7 Code privately
No prompt logging. No data used for training.