Moonshot AI releases Kimi K2.7 Code with 1T parameters, 256K context window, 30% lower thinking token usage
Moonshot AI has released Kimi K2.7 Code, a 1 trillion parameter Mixture-of-Experts model designed for long-horizon coding tasks. The model features a 256K context window and reduces thinking token usage by approximately 30% compared to its predecessor K2.6.
Kimi K2.7 Code — Quick Specs
Moonshot AI Releases Kimi K2.7 Code with 1T Parameters
Moonshot AI has released Kimi K2.7 Code, a 1 trillion parameter Mixture-of-Experts (MoE) model built on the K2.6 architecture. The model features 32 billion activated parameters, a 256K token context window, and reduces thinking token usage by approximately 30% compared to K2.6.
Architecture and Specifications
Kimi K2.7 Code uses a 384-expert MoE architecture with 8 experts selected per token, plus 1 shared expert. The model includes 61 total layers (including 1 dense layer), 64 attention heads, and a 160K vocabulary size. It employs Multi-head Latent Attention (MLA) with SwiGLU activation and integrates MoonViT, a 400M parameter vision encoder, for multimodal capabilities supporting image and video input.
The model is available in native INT4 quantization using the same method as Kimi-K2-Thinking.
Benchmark Performance
According to Moonshot AI, K2.7 Code shows substantial improvements across coding and agentic benchmarks:
Coding benchmarks:
- Kimi Code Bench v2: 62.0 (vs 50.9 for K2.6)
- Program Bench: 53.6 (vs 48.3 for K2.6)
- MLS Bench Lite: 35.1 (vs 26.7 for K2.6)
Agentic benchmarks:
- Kimi Claw 24/7 Bench: 46.9 (vs 42.9 for K2.6)
- MCP Atlas: 76.0 (vs 69.4 for K2.6)
- MCP Mark Verified: 81.1 (vs 72.8 for K2.6)
Moonshot AI compared K2.7 Code against GPT-5.5 and Claude Opus 4.8, though these comparison scores cannot be independently verified. In Moonshot's testing, GPT-5.5 led most benchmarks, with K2.7 Code placing second or third depending on the task.
Deployment and Availability
The model API is available at platform.moonshot.ai with OpenAI and Anthropic-compatible interfaces. Pricing per million tokens has not been disclosed. The model requires transformers version 4.57.1 or higher (but below 5.0.0) and can be deployed using vLLM, SGLang, or KTransformers inference engines.
Kimi K2.7 Code operates exclusively in thinking mode with preserve_thinking forced to True. Moonshot AI recommends a temperature of 1.0 and top_p of 0.95 for inference. Video content chat is currently experimental and only supported through the official API.
What This Means
Kimi K2.7 Code represents Moonshot AI's push into specialized coding models with extended reasoning capabilities. The 30% reduction in thinking token usage addresses a practical efficiency concern for reasoning models in production environments. The 256K context window positions it competitively for repository-level code understanding tasks, though it remains shorter than some competitors offering 1M+ token windows. The model's MoE architecture with 384 experts and INT4 quantization support suggests Moonshot is optimizing for deployment efficiency alongside raw capability.
Related Articles
DeepSeek Launches V4.1 Flash: Low-Cost MoE Model Claims to Beat V4 Pro
DeepSeek has released V4.1 Flash, a sparse mixture-of-experts model priced at $0.30 per 1M input tokens and $1.20 per 1M output tokens with a 1 million token context window. DeepSeek claims the model exceeds the larger V4 Pro on performance, speed, and task completion time.
Alibaba Releases Qwen3.8 Max (0902), a 2.4-Trillion-Parameter MoE Model With 1M-Token Context
Alibaba's Qwen team released Qwen3.8 Max (0902), a 2.4-trillion-parameter mixture-of-experts model with a 1M-token context window that accepts text, image, and video input. The snapshot is post-trained for coding, agentic workflows, and long-horizon task execution, priced at $2/$6 per 1M input/output tokens.
DeepSeek V4.1-Flash Cuts KV Cache Memory by Up to 8x, Matches Opus 5 on Coding Benchmark
DeepSeek released V4.1-Flash, a 552-billion-parameter model built to slash the memory overhead of long-context AI agents. The model cuts GPU cache needs to roughly a quarter of its predecessor's and matches closed models from OpenAI and Anthropic on select coding benchmarks.
DeepSeek Releases V4.1-Flash: 552B MoE Model Cuts KV Cache to 890 Bytes Per Token
DeepSeek has released V4.1-Flash, a 552B-parameter multimodal Mixture-of-Experts model supporting 1M-token context and activating only 8B parameters during prefill. The model uses a new Causal Encoder-Decoder architecture and Compressed Sparse Attention 2 to cut global KV cache to 890 bytes per token, roughly a quarter of its predecessor.
Comments
Loading...