Breaking

Moonshot AI makes Kimi K2.7 Code HighSpeed generally available

July 10, 2026 at 10:00 EDT

Software Dev & Coding
Foundation Models
AI Agents

On July 10, 2026, Moonshot AI began general availability of "K2.7 Code HighSpeed," the faster tier of its coding-focused model "Kimi K2.7 Code." After roughly a three-week beta, it became available on paid plans from Allegretto and above, alongside a Vercel Plugin for the Kimi Code CLI, an improved Web UI, and Archived Sessions for viewing and restoring past sessions.

July 10, 2026 · Moonshot AI

Kimi K2.7 Code HighSpeed Goes Live — Speed for a Price

The faster serving tier of the coding-focused K2.7 Code model reaches general availability, delivering roughly 5–6× the standard generation speed to keep agentic-coding developers in flow — at about twice the cost.

~260

tokens/s peak in short contexts (~180 t/s median)

5–6×

faster generation vs the standard version

2×

the API price of the standard tier

262K

token context window (both tiers)

Speed vs. Cost — the trade-off, drawn to scale

Output token pricing ($/M). Taller stack = higher cost.

$4.00

Standard

$8.00

HighSpeed

Same model, same 262K context — HighSpeed only changes the serving tier to prioritize output speed.

K2.7 Code vs K2.6 — benchmark gains

Score improvements powering the faster coding agent.

50.9

62.0

Kimi Code Bench v2

+21.8%

48.3

53.6

Program Bench

+11.0%

K2.6 K2.7 Code

Open-weight MoE design: ~1T total / ~32B active params, ~30% fewer thinking tokens than the prior model.

Developers in favor

"2.5–4.2× faster with comparable quality." "260 tokens/s is practical" for keeping agentic workflows uninterrupted.

Words of caution

Higher credit consumption — "use only for tasks that truly need the speed." Some note the absence of independent benchmark submissions.

Shipping alongside HighSpeed

Vercel Plugin for Kimi Code CLI Enhanced Web UI Archived Sessions Toggle standard ⇄ HighSpeed

Available on paid plans from Allegretto and above. Built for long-horizon agent tasks, debugging, large codebase exploration, and Vercel deployment integration.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.