Moonshot AI's 2026-06 coding-specialist model (1T MoE, 32B active). 256K context. +21.8% on Kimi Code Bench v2 vs K2.6, 30% fewer reasoning tokens. Open-weight (Modified MIT).
Kimi K2.7 Code is Moonshot AI's June 2026 coding-specialist member of the Kimi K2 family. It keeps the 1T-parameter Mixture-of-Experts shape with 32B active parameters per token and targets agentic software engineering workloads: multi-file edits, MCP tool invocation, repository analysis, and long-horizon implementation loops.
For TheRouter customers, the important production shape is narrower than Moonshot's own platform story: this route is cataloged as text input to text output, with 256K context and 65,536 maximum completion tokens from standard-models.yaml. Use it when coding quality and token efficiency matter more than a no-thinking fast lane; use a separate low-latency route for trivial tool dispatch or interactive chat.
Moonshot documents Kimi K2.7 Code as a Kimi-platform coding model with mandatory thinking, published pricing, and discussion of the wider K2.7 family including high-speed routing options.
TheRouter currently publishes moonshot/kimi-k2.7-code as a text-to-text chat-completions route with 256K context, 65,536 max completion tokens, and TheRouter platform pricing from standard-models.yaml.
| Type | Rate |
|---|---|
| Input | $1.03 / 1M tokens |
| Output | $4.32 / 1M tokens |
| Release date | 2026-06-12platform.kimi.ai ↗ | verified |
| Architecture | 1T-total / 32B-active MoE; 384 experts, 8 routed plus 1 shared expert per forward passplatform.kimi.ai ↗ | verified |
| Context window | 256,000 tokensplatform.kimi.ai ↗ | verified |
| Thinking mode | Always on; non-thinking mode is not supportedplatform.kimi.ai ↗ | verified |
| Training data cutoff | Not publicly disclosed | unknown |
| License | Modified MIT Licenseplatform.kimi.ai ↗ | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Kimi Code Bench v2 Moonshot-reported score; +21.8% versus K2.6. | 62.0% | platform.kimi.ai ↗ | |
Program Bench Moonshot-reported score; +11.0% versus K2.6. | 53.6% | platform.kimi.ai ↗ | |
MCP Mark Verified Moonshot-reported MCP tool invocation benchmark. | 81.1% | platform.kimi.ai ↗ | |
Kimi Claw 24/7 Moonshot-reported sustained agent benchmark; +9.3% versus K2.6. | 46.9% | platform.kimi.ai ↗ |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "moonshot/kimi-k2.7-code",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Use the standard TheRouter chat-completions endpoint for coding prompts.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"moonshot/kimi-k2.7-code","messages":[{"role":"user","content":"Review this diff for regressions."}]}'TheRouter's June note highlighted the two routing consequences of K2.7 Code: lower reasoning-token burn for standard coding agents, and a separate HighSpeed sibling for latency-sensitive work.
The launch coverage recorded the operator constraint that matters most for routing: K2.7 Code does not offer a non-thinking mode, so cost and latency models must include reasoning overhead.
No. It is a strong route for deep coding-agent work, but mandatory thinking makes it inefficient for trivial tool dispatch, quick formatting, or latency-sensitive chat. Keep a cheaper non-thinking route for those turns.
No. standard-models.yaml lists moonshot/kimi-k2.7-code as text input to text output. If the request contains screenshots, route to a cataloged multimodal model such as moonshot/kimi-k2.6.
The live pricing displayed on the model card comes from TheRouter's standard-models.yaml, not from prose in this page. At curation time the catalog lists $0.95 per million input tokens and $4.00 per million output tokens.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | platform.kimi.ai ↗ | 2026-08-04 | verified |
| Architecture | platform.kimi.ai ↗ | 2026-08-04 | verified |
| Context window | platform.kimi.ai ↗ | 2026-08-04 | verified |
| Thinking mode | platform.kimi.ai ↗ | 2026-08-04 | verified |
| Training data cutoff | — | — | unknown |
| License | platform.kimi.ai ↗ | 2026-08-04 | verified |
| Kimi Code Bench v2 | platform.kimi.ai ↗ | 2026-08-04 | verified |
| Program Bench | platform.kimi.ai ↗ | 2026-08-04 | verified |
| MCP Mark Verified | platform.kimi.ai ↗ | 2026-08-04 | verified |
| Kimi Claw 24/7 | platform.kimi.ai ↗ | 2026-08-04 | verified |
| Kimi K2.7 Code API: 30% Fewer Thinking Tokens and a HighSpeed Variant Change Your Routing Policy | Kimi Open Platform ↗ | 2026-06-17 | verified |
| Kimi K2.7 Code API: Thinking-Only Constraint Changes How You Route Coding Agents | Moonshot AI ↗ | 2026-06-13 | verified |
| Is moonshot/kimi-k2.7-code the right default for every coding request? | platform.kimi.ai ↗ | 2026-08-04 | to verify |
| Does this TheRouter route support image input? | platform.kimi.ai ↗ | 2026-08-04 | to verify |
| How is K2.7 Code priced on TheRouter? | platform.kimi.ai ↗ | 2026-08-04 | to verify |