Friday, July 17, 2026Scoutari
Log inSign up
Sign up
ModelsSkill RadarSavedPersonalizeSubmit sourceLog in
Back to timeline

Thu, July 1613:44Model/APIModel releasesAPI & pricingChinese modelsModel releases guide

Moonshot AI Releases Kimi K3: 2.8T Parameters & 1M Context

View originalOpen Kimi

Decision Brief

What changedMoonshot AI released Kimi K3, an open-source model with 2.8T parameters, 1M token context, and native multimodal input.
Why it mattersKimi Delta Attention boosts decoding speed 6.3x, Attention Residuals improve training efficiency 25%, a boon for developers needing long context and efficient inference.
Who should careTeams building on model APIs
Affected stackClaudeOpenAIKimi
Builder actionWorth evaluating: can Claude replace or complement your current model
Source confidenceMedium · Reliable media or first-hand reporting

Moonshot AI launched Kimi K3, a frontier-level open-weight model with 2.8T parameters, 1M token context window, and native multimodal input. It introduces Kimi Delta Attention (KDA) for up to 6.3x faster decoding and Attention Residuals for ~25% training efficiency gains. K3 is available on multiple platforms, with open weights promised by July 27, 2026. In Frontend Code Arena, K3 leads with a 76% pairwise win rate, surpassing Claude Fable 5 and GPT-5.6 Sol in several benchmarks, but lags behind in overall user experience. Independent evaluations rate it comparable to Opus 4.8 and GPT-5.5. This release marks a major milestone for open-source models, offering significant speed and efficiency gains for long-context or multimodal tasks, though overall experience needs refinement.

Summary basis: official / RSS sourceCompiled from the source scope noted above; the original remains authoritative.

Sources

  • AINews（smol.ai）

    Weekday recap of top news for AI engineers — dense, builder-oriented signal.

  • AINews（smol.ai）

Related intel

留言

登入后即可留言，和其他 builder 交换实测心得。

还没有留言，抢头香。