Moonshot quietly ships Kimi K2.8 Preview in Kimi Code: multimodal, 1M context for all tiers, 'close to K3'
On Sept 11, 2026 Moonshot AI replaced the model behind its Kimi Code coding tool with Kimi K2.8 Preview, a mid-tier model between Kimi K2.7 Code and the flagship Kimi K3. It adds image and video input, gives every membership tier a 1M-token context, and offers K3-style low/high/max thinking effort (max by default). Moonshot says it performs close to K3 with higher thinking efficiency. The model id stayed `kimi-for-coding`, and no benchmarks, parameter count or per-token price were published.
Key facts
- Rolled out Sept 11, 2026 on Kimi Code under the unchanged model id kimi-for-coding (per the Kimi Code changelog, as reported)
- Context up to 1,048,576 tokens on all membership tiers; image and video input
- Thinking effort low / high / max (default max)
- Moonshot: overall performance 'close to K3' with higher thinking efficiency (no benchmark table)
- Not listed on Moonshot's open-platform API model list at launch; kimi-for-coding-highspeed still served K2.7 Code HighSpeed
What happened
A silent backend upgrade to Moonshot's coding subscription product, released ahead of the expected Kimi K3.1.
Why it matters
It is a small step in Moonshot's fast release cadence. Shipping a model without changing its model id makes it hard for users to tell which model they are actually calling.
Unverified: Moonshot's own changelog page was not read directly; facts come from secondary reports quoting it.
Changelog
- 2026-10-01: created (leads run, missed pre-window item)
Related events
Sources (3)
- pressIntelligent Living: Kimi K2.8 Preview, 1M context behind one unchanged model ID
- pressPhemex News: Moonshot AI launches Kimi K2.8 Preview on Kimi Code
- pressEmergent: Moonshot AI launches Kimi K2.8 Preview multimodal model
id: 2026-09-11-kimi-k2-8-preview · updated 2026-10-01 · open in the interactive timeline