Kimi K2.7 Code
Kimi K2.7 Code by Kimi: thinking-only agentic coding MoE, 1T params (32B active), 256K context, native multimodality, 30% fewer thinking tokens than K2.6.
Model Overview
Capabilities, design details, and architectural traits
Kimi K2.7 Code - Thinking-only agentic coding MoE with reduced overthinking
Kimi K2.7 Code is Kimi's dedicated coding model, built on K2.6 with a focus on long-horizon coding task completion and instruction compliance in extended contexts. Its defining characteristic is that it operates exclusively in thinking mode - non-thinking mode is unsupported and disabling it causes an API error.
The model reduces overthinking tendencies by 30% on average compared to K2.6, using fewer thinking tokens while improving task success rates. It also improves agentic capabilities by 10% over K2.6.
| Trait | Detail |
|---|---|
| Thinking-only mode | No non-thinking mode available; disabling thinking causes an API error |
| Reduced overthinking | ~30% fewer thinking tokens than K2.6 on average |
| Architecture | 1T-parameter MoE with 32B active, 384 experts, 8 selected per token, 1 shared expert |
| Native INT4 quantization | MoE weights stored in INT4; BF16 for non-MoE layers |
| Multimodal input | MoonViT vision encoder supports text, image, and video input |
| Fixed sampling | temperature locked at 1.0, top_p at 0.95, n at 1, penalties at 0.0 |
| HighSpeed variant | Same model at ~180 tokens/s, up to 260 tokens/s in short contexts |
| Context window | 256K tokens |
Long-horizon coding and agentic tool use
Kimi K2.7 Code supports multi-step tool invocation and reasoning, combining visual understanding with function calling. The model targets complex logical reasoning, mathematical problems, and code writing within its 256K context window.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Kimi | Anthropic | Anthropic |
| Release Date | June 12, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 262K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.95 Best Input Pricing | $4 | $2 |
| Output Pricing | $4 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | textimagevideo | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 25.8 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 60.8 | - | - |
| Agentic Index | 21.0 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Kimi
Other models by Kimi
Top AI Models
Leading alternatives by intelligence score
