Kimi K2.8 is live in Kimi Code, with 1M context on every tier
Moonshot's mid-tier coder now sits between K2.7 Code and flagship K3 — cheaper calls, adjustable thinking, but closed weights and embargoed specs.
Copy markdown
Now fully live across Kimi Code and Work
Kimi K2.8 Preview is rolled out to every Kimi Code and Kimi Work user with no config change — a mid-tier step between K2.7 Code and the flagship K3, with 'thinking' you can turn up or down per call. The API id stays kimi-for-coding, so existing setups pick it up automatically.
1M context on every tier, not just the top plan
The window jumps to 1,000,000 tokens across all membership tiers — up from 262,144 on K2.7 Code, roughly 4x more room for whole-repo prompts or long agent runs without upgrading to a premium plan.
What it costs to call
Moonshot hasn't posted official pricing yet; through gateways it's landing near $0.80 per million input tokens and $3.35 output, with cached reads at $0.14 — mid-tier rates for coding Moonshot claims is close to K3.
The catch: closed weights, embargoed specs
Unlike its open-weight siblings K2.7 Code and K3, K2.8 ships proprietary — no download, no parameter count, and 'close to K3' is Moonshot's own number with no third-party benchmarks yet. It's a hosted-only bet until the preview ends.
Elsewhere: Anthropic's new threat report targets your API keys
Anthropic's September threat intelligence details real misuse that ran on agent frameworks and hunted cached API keys — worth a read for anyone shipping agents, and a nudge to scope tokens tight and rotate them.