AI
Kimi K3, and what we can still learn from the pelican benchmark
Chinese AI lab Moonshot AI has announced Kimi K3, a 2.8 trillion parameter model.
Key takeaways
- Kimi K3 is Moonshot AI's most capable model to date, with 2.8 trillion parameters.
- Kimi K3 reaches an overall Elo of 1547 on Artificial Analysis's private long-horizon knowledge work evaluation.
- Kimi K3's pricing of $3/million input and $15/million output tokens makes it the most expensive model released by a Chinese AI lab.
- The pelican benchmark has mostly lost its correlation to how good models actually are.
Chinese AI lab Moonshot AI has announced Kimi K3, a 2.8 trillion parameter model. Currently available via website and API, its open weights are scheduled for release by July 27, 2026. Priced at $3 per million input tokens and $15 per million output tokens, it is the most expensive model released by a Chinese AI lab to date. On Artificial Analysis's private long-horizon knowledge work evaluation, Kimi K3 achieved an overall Elo of 1547, placing it behind only Claude Fable 5.
In their words
“On our private long-horizon knowledge work evaluation, Kimi K3 reaches an overall Elo of 1547, +732 points from Kimi K2.6 and behind only Claude Fable 5.”
“Cost per task ($0.94) is similar to GPT-5.6 Sol ($1.04), ~1/2 the price of Opus 4.8 ($1.80) and higher than open weights peers”
By the numbers
- 2.8T
- Parameters in the Kimi K3 model
- +732
- Elo point increase of Kimi K3 over Kimi K2.6
- $3
- Cost per million input tokens for Kimi K3
- $15
- Cost per million output tokens for Kimi K3
- -21%
- Decrease in output token usage compared to K2.6
How it unfolded
- Moonshot AI announces Kimi K3 model
- Promised open weight release of Kimi K3
Turn stories like this into views
Ravenclip finds the AI news, makes the video, and posts it before attention moves on.
Common questions
- What happened with Kimi K3?
- Kimi K3 is Moonshot AI's most capable model to date, with 2.8 trillion parameters.
- Where can I read the original report?
- Read the full report at simon_willison.
