Kimi For Coding

The kimi-k2.org status tracker, updated July 16, 2026, reports that official Kimi Code model configuration documentation lists Kimi K3 as available under model ID k3, sitting alongside two K2.7 Code IDs: kimi-for-coding and kimi-for-coding-highspeed. This clarifies the current model lineup exposed by the Kimi For Codin The page also documents plan-gated access that developers need to know when integrating against api.kimi.com/coding/v1: the Andante tier does not support k3, Moderato supports k3 up to 256k context, and Allegretto and above unlock the full 1M context, while K2.7 Code IDs remain available across tiers and reasoning effo

According to a HowAIWorks.ai model profile dated July 16, 2026, Moonshot AI's Kimi K3 is documented as the flagship model available through Kimi Code and the Kimi API, announced that day and displacing K2.6 in the official Kimi API documentation. It is described as a 2.8-trillion-parameter Mixture-of-Experts model usin The same profile documents developer-relevant changes for anyone integrating via the Kimi For Coding provider: the context window expands to 1M tokens (4x K2.6's 256K) with flat $3/$15 per million token pricing regardless of context length, and reasoning is always on, currently locked to max effort. Moonshot's own 35-b

The Kimi Code CLI changelog (kimi.com) provides the authoritative release notes for the tooling that backs the Kimi For Coding provider. Version 0.43.1 (2026-09-15) adds native clipboard support on Linux X11 so copying from the TUI no longer depends on terminal OSC 52 support, reduces event-loop stalls and GC churn in Version 0.43.0 (2026-09-14) is documented as a feature release with Web AI session titles on by default (regenerable from the rename field), Ctrl-X session deletion in the session picker with a y confirmation prompt, a new -y / --yes flag on `kimi upgrade` (alias `kimi update`) to install updates directly, and a `loop

The upstream Kimi CLI (moonshotai/kimi-cli) shipped v1.50.0 on 2026-09-02 with a targeted Kosong fix: empty anthropic-beta headers are no longer sent when no beta features are declared, because adaptive thinking removes the interleaved-thinking beta and some backends were rejecting requests with an empty header value. v1.49.0 (2026-07-16) changes how Kimi providers size completions: the completion-token budget is now clamped to the model's remaining context window instead of a fixed max tokens=32000, with an explicit hard cap available via KIMI MODEL MAX COMPLETION TOKENS (KIMI MODEL MAX TOKENS remains a compatibility alias; 0 or ne

Kimi Code's official "What's New" page documents two dated developments relevant to the Kimi For Coding provider. The first is the full rollout of K2.8 Preview on September 11, 2026 behind the unchanged Model ID kimi-for-coding, so clients and third-party tools need no configuration changes; Moonshot describes its codi The same page highlights Kimi Code CLI v0.43.0 (September 14, 2026) as a separate product update for the provider's tooling, making AI session titles on by default in Web, adding session deletion via Ctrl-X then y in the session selector, and removing the 24-hour cap from Goal Mode time budgets so closed-session time n