Moonshot AI Kimi K2.7 Code Highspeed — accelerated inference variant of K2.7 Code. Same coding-specialist capabilities with higher throughput. 256K context.
Kimi K2.7 Code Highspeed is Moonshot AI's accelerated inference SKU for the K2.7 Code coding-specialist model. It keeps the same text-only coding-agent profile as K2.7 Code on TheRouter — 256K context, 65,536 maximum completion tokens, tool calling, and structured-output support — but prices and routes it as the latency-oriented sibling.
Use this route when K2.7 Code quality is needed but interactive agent latency is the bottleneck. The tradeoff is explicit: TheRouter's catalog prices Highspeed at 2× the standard K2.7 Code rate, so it should be reserved for human-in-the-loop coding, fast review loops, and agent orchestration where seconds saved matter more than unit cost.
Moonshot presents Highspeed as part of the Kimi K2.7 Code platform family for faster coding-agent inference.
TheRouter exposes moonshot/kimi-k2.7-code-highspeed as its own text-to-text chat-completions model id with 256K context, 65,536 max completion tokens, and 2× catalog pricing versus moonshot/kimi-k2.7-code.
| Type | Rate |
|---|---|
| Input | $2.05 / 1M tokens |
| Output | $8.64 / 1M tokens |
Moonshot prices -highspeed as a separate SKU at 2x the standard K2.7-code rate (platform.kimi.ai/docs/pricing/chat-k27-code, international USD card, read 2026-07-29).
| Release date | 2026-06-12platform.kimi.ai ↗ | verified |
| Model family | Kimi K2.7 Code accelerated Highspeed variantplatform.kimi.ai ↗ | verified |
| Context window | 256,000 tokensplatform.kimi.ai ↗ | verified |
| Maximum completion | 65,536 tokensplatform.kimi.ai ↗ | verified |
| Catalog pricing | 2× standard K2.7 Code rate on TheRouterplatform.kimi.ai ↗ | verified |
| Training data cutoff | Not publicly disclosed | unknown |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Kimi Code Bench v2 Same model-quality basis as K2.7 Code; Highspeed is an inference SKU, not a separate benchmark release. | 62.0% | platform.kimi.ai ↗ | |
Program Bench Moonshot reports this for K2.7 Code; use it for quality comparison, then use Highspeed only when latency matters. | 53.6% | platform.kimi.ai ↗ | |
MCP Mark Verified Moonshot-reported MCP tool invocation score for the K2.7 Code model family. | 81.1% | platform.kimi.ai ↗ | |
Highspeed throughput target Throughput target for the accelerated Highspeed route; compare cost-normalized latency, not only benchmark quality. | 180 tokens/sectokens/sec | 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-highspeed",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Use the dedicated Highspeed model id when latency justifies the higher price.
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-highspeed","messages":[{"role":"user","content":"Review this patch for release-blocking bugs."}]}'The routing implication is simple: standard K2.7 Code should carry ordinary deep coding turns, while Highspeed should be reserved for latency-sensitive loops where doubled unit price is justified.
Choose Highspeed when a human or orchestrator is waiting on the coding-agent result and lower latency changes the experience. For background analysis, scheduled refactors, and batch generation, use the standard K2.7 Code route to avoid paying the doubled rate.
No. TheRouter catalog data for moonshot/kimi-k2.7-code-highspeed keeps 256K context and 65,536 maximum completion tokens. The difference is routing, throughput, and price rather than a larger memory window.
No. TheRouter lists moonshot/kimi-k2.7-code-highspeed as text input to text output. Route screenshots, UI images, or video understanding to a cataloged multimodal Moonshot model such as moonshot/kimi-k2.6.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | platform.kimi.ai ↗ | 2026-08-14 | verified |
| Model family | platform.kimi.ai ↗ | 2026-08-14 | verified |
| Context window | platform.kimi.ai ↗ | 2026-08-14 | verified |
| Maximum completion | platform.kimi.ai ↗ | 2026-08-14 | verified |
| Catalog pricing | platform.kimi.ai ↗ | 2026-08-14 | verified |
| Training data cutoff | — | — | unknown |
| Kimi Code Bench v2 | platform.kimi.ai ↗ | 2026-08-14 | verified |
| Program Bench | platform.kimi.ai ↗ | 2026-08-14 | verified |
| MCP Mark Verified | platform.kimi.ai ↗ | 2026-08-14 | verified |
| Highspeed throughput target | platform.kimi.ai ↗ | 2026-08-14 | verified |
| Kimi K2.7 Code API: 30% Fewer Thinking Tokens and a HighSpeed Variant Change Your Routing Policy | Kimi Open Platform ↗ | 2026-08-14 | verified |
| When should I choose Highspeed instead of standard K2.7 Code? | platform.kimi.ai ↗ | 2026-08-14 | to verify |
| Does Highspeed change the model's context window or output limit? | platform.kimi.ai ↗ | 2026-08-14 | to verify |
| Is this route multimodal? | platform.kimi.ai ↗ | 2026-08-14 | to verify |