Kimi K2.7 Code HighSpeed review, pricing and limits

K2.7 Code HighSpeed fits teams running autonomous coding agents with heavy tool use who want lower latency without an audited SWE-bench score to justify the pick. It cuts reasoning-token usage roughly 30% versus the K2.6 base model, lowering real-world cost on long agentic sessions, though Moonshot itself still recommends K2.6 for general-purpose writing work.

Kimi K2.7 Code HighSpeed scores 81.1 on MCP Mark Verified, beating Claude Opus 4.8's 76.4 on the same tool-calling test. It's Moonshot AI's fast-serving mode for its 1-trillion-parameter open-weight coding model, reaching roughly six times the throughput of standard K2.7 Code on coding completions.

Provider: Moonshot AI · Family: Kimi K2

More about Moonshot AI on HokAI

Context window: 262,144 tokens · Max output: 49,152

Input modalities: text, image, video, tool-calls, code · Output: text, tool-calls, code

About Kimi K2.7 Code HighSpeed

Kimi K2.7 Code HighSpeed is a fast-inference variant of Kimi K2.7 Code, the coding-focused release Moonshot AI shipped on June 12, 2026 as a fine-tune of its Kimi K2.6 base. The HighSpeed mode itself followed three days later, on June 15, 2026, rolling out first to the Kimi Code Beta channel. Architecturally it is unchanged from K2.6: a Mixture-of-Experts transformer with 1 trillion total parameters and 32 billion activated per token, 61 layers, 384 experts with 8 routed plus 1 shared expert per token, Multi-head Latent Attention (MLA) for KV cache compression, SwiGLU activations, and a 160K vocabulary. A MoonViT vision encoder gives it native image and video input alongside text, and the context window carries over unchanged from K2.6 at 256K (262,144) tokens.

Moonshot has not submitted K2.7 Code to any independently audited benchmark suite as of late June 2026. The only numbers on the table are Moonshot's own Kimi Code Bench v2, where K2.7 Code scores 62.0 versus K2.6's 50.9 (a 21.8% relative gain), Program Bench at 53.6 versus 48.3 (+11.0%), and MLS Bench Lite at 35.1 versus 26.7 (+31.5%). On MCP Mark Verified, a third-party tool-invocation benchmark covering Notion, GitHub, Postgres, Filesystem, and Playwright environments, K2.7 Code scores 81.1, ahead of Claude Opus 4.8's 76.4 on that specific test, though Opus 4.8 leads on broader coding suites like Terminal-Bench 2.1. Its closest open-weight rival, GLM-5.2, holds a stronger position on audited benchmarks at 62.1% on SWE-bench Pro, so K2.7 Code's real edge is token efficiency and tool-call reliability rather than raw solve rate.

The defining change from K2.6 is a roughly 30% cut in reasoning-token usage per task, which lowers effective cost on reasoning-heavy coding workloads even though the per-token price is nearly identical. K2.7 Code always runs with extended thinking enabled; a request sent with thinking explicitly disabled is silently rerouted to K2.6 rather than erroring; interleaved thinking is preserved across multi-turn tool-calling sessions for coherent long-horizon coding runs.

The HighSpeed mode is a serving-side optimization, not a retrained model: Moonshot reports substantially faster output on median-length coding completions and bursts even higher on short-context requests, roughly six times the throughput of the standard K2.7 Code deployment. Third-party inference provider Crusoe reports over 430 output tokens/sec running K2.7 on its own optimized stack, which shows the ceiling is provider-dependent rather than fixed by the model.

K2.7 Code HighSpeed is priced per token on Moonshot's native Kimi API, with a cheaper rate available through OpenRouter. The model is also live on Cloudflare Workers AI and the Vercel AI Gateway. Exact rates and self-hosting cost comparisons are broken out below.

Weights are published under a Modified MIT License on Hugging Face and GitHub: free for research and most commercial use, with an attribution clause that kicks in for very large-scale commercial deployments, the same style of condition Moonshot has used since the original K2 release. Self-hosting the FP8 weights needs roughly 1TB of HBM, realistically an 8x H200 SXM5 node, leaving enough headroom for KV cache at full context with small batches. Moonshot also ships native INT4 weights trained with quantization-aware training rather than post-hoc quantization, cutting VRAM roughly in half with minimal quality loss; vLLM, SGLang, and KTransformers are the officially recommended INT4 serving engines.

An independent academic evaluation of the prior K2.5 model found it had a lower overrefusal rate and higher willingness to cooperate with borderline requests than GPT-5.2 and Claude Opus 4.5, a pattern that may carry over until Moonshot publishes its own K2.7 assessment. Training data cutoff is unconfirmed for K2.7 Code specifically; K2.6's was reported at approximately April 2025, though Moonshot never verified that figure itself.

Pricing

Moonshot's native Kimi API prices HighSpeed mode at $0.95 per 1M input tokens and $4.00 per 1M output tokens, with cached input discounted to $0.19 per 1M. OpenRouter carries the same model cheaper, at $0.74 input and $3.50 output per 1M tokens. There's no subscription tier: self-hosting the open weights avoids per-token billing but needs roughly 1TB of GPU memory for the FP8 build.

What a real job costs

Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.

Key Features

• HighSpeed serving mode: Reaches roughly 180 tokens/sec on median coding completions and up to 260 tokens/sec on short bursts, about 6x the throughput of standard K2.7 Code.
• 262,144-token context window: Compressed with Multi-head Latent Attention (MLA) so self-hosted deployments can keep enough KV cache headroom for long agentic sessions.
• Always-on interleaved thinking: Extended thinking is always active and carries over between tool calls in a single session, so multi-step coding tasks don't lose their reasoning thread.
• Native MoonViT vision encoder: A 400M-parameter encoder handles image and video input natively, without a separate captioning step before the coding model sees them.
• Native INT4 quantization: Quantization-aware-trained INT4 weights ship alongside FP8, roughly halving self-hosting VRAM with vLLM, SGLang, and KTransformers serving support.

Pros

• Its MCP tool-invocation reliability held up across five different environments in the same third-party benchmark, not just a single integration test.
• The Modified MIT License permits most commercial use with no subscription fee at all, a real advantage against closed, API-only coding models.
• Cuts reasoning-token overhead enough to matter on long agentic coding sessions, without raising the per-token price much over the previous generation.

Cons

• SWE-bench Verified, SWE-bench Pro, LiveCodeBench, and GPQA Diamond scores are all missing for this specific release, only Moonshot's own in-house numbers are public.
• No system card has been published for this release, so refusal policy, jailbreak resistance, and agentic-misuse evaluations are all undocumented.
• HighSpeed mode is beta-gated to the Kimi Code Beta channel and may not be available on every account.

Benchmarks

• Program Bench: 53.6 vendor-reported · 12 Jun 2026
• Mls Bench Lite: 35.1 vendor-reported · 12 Jun 2026
• Mcp Mark Verified: 81.1 vendor-reported · 12 Jun 2026
• Kimi Code Bench V2: 62 vendor-reported · 12 Jun 2026

A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.

Visit Kimi K2.7 Code HighSpeed Official Page