Moonshot quietly rolls out Kimi K2.8 Preview on Kimi Code: 1M context unlocked for every membership tier

On September 11, 2026, Moonshot AI silently rolled out Kimi K2.8 Preview across Kimi Code, with Kimi Work following at the same time. The defining detail of the release is that the model ID did not change: it is still kimi-for-coding, so Kimi Code clients and any third-party tool plugged into that ID need no configuration changes and traffic flows to the new model automatically. Moonshot positions the preview as close to flagship K3 in overall capability, with notably better thinking efficiency than K2.7 Code and broad improvements in coding and agent tasks. The confirmed specifics are: low/high/max thinking-effort tiers aligned with K3 and defaulting to max; a 1,048,576-token context window unlocked for every membership tier; and routing both K3-series and K2.8 Preview requests to the no-thinking K2.8 Preview when thinking is disabled. The change lowers the daily-use floor substantially: previously the 1M context was available only to Allegretto (199 RMB/month) and above, but K2.8 Preview brings 1M down to the Adagio free tier. The move addresses commercial pressure that has built since August 2026. Bloomberg reports that Moonshot's ARR was about $100M in March, $200M in April, above $300M in June, and crossed $1B in August, with an internal target of $2B by year-end. K3 was a key inflection point: released in July with open weights, ultra-long context, and flagship coding performance, it topped the Arena.ai frontend coding leaderboard with 1679 points, ahead of Claude Fable 5 and GPT-5.6 Sol. Separately, Bloomberg reports that Moonshot submitted an A1 listing application to the Hong Kong Stock Exchange confidentially earlier this month, formally kicking off its Hong Kong IPO process. Valuation history: $4.3B post-money after Series C at end-2025, $20B in May, $35B after the Series F in July, and a $50B pre-money mark for the Pre-IPO round.