AI is changing the world. Don't stay behind. Clear summaries, community insight, delivered without the noise. Subscribe to never miss a beat.

Privacy

Contact

Back to all news

kimi-k3 llms moonshot-ai ai-benchmarks

Kimi K3, and what we can still learn from the pelican benchmark

simonwillison.net

July 17, 2026

6 min read

🔥🔥🔥🔥🔥

50/100

Summary

Moonshot AI has launched Kimi K3, their most capable model to date with 2.8 trillion parameters. It is currently accessible via their website and API, with an open weight release scheduled for July 27, 2026.

Key Takeaways

Moonshot AI announced the release of Kimi K3, their most capable model to date, featuring 2.8 trillion parameters.
Kimi K3 is the first "open 3T-class model" and is currently available via their website and API, with an open weight release expected by July 27, 2026.
Kimi K3 achieves an overall Elo of 1547 in long-horizon knowledge work evaluation, significantly outperforming its predecessor Kimi K2.6.
The pricing for Kimi K3 is set at $3 per million input tokens and $15 per million output tokens, making it the most expensive model released by a Chinese AI lab to date.

Read original article

Community Sentiment

Mixed

Positives

The gap is closing — Kimi 3 is only 3 months behind the US model, signaling progress in AI capabilities.
Fable still reigns supreme for complex animation tasks, showcasing the high bar for creativity in AI models.
Gemini 3.1 Pro is noted for its visual creativity, despite some limitations, indicating strong potential in AI design.

Concerns

Benchmarks are losing credibility; experienced users express skepticism about their relevance for frontier models.
Concerns arise that if pelican SVGs were part of the training data, the model should perform at a higher artistic level.
Some commenters doubt the utility of benchmarks altogether, suggesting that models like Sol lack the creativity and 'taste' expected from advanced AI.