China's Moonshot AI has taken the top spot in the Frontend Code Arena with its model Kimi-K3, leapfrogging Anthropic's Claude Fable 5. Kimi-K3 scored 1,679 points, a 48-point lead over Fable 5's 1,631. The jump was dramatic — from 18th place with the previous version, Kimi-K2.6, straight to first.
The 17-Slot Climb
Kimi-K3 didn't just edge past its rivals. It vaulted 17 positions in a single update. The 48-point gap between first and second is more than three times the 13-point margin separating Fable 5 from third-place GPT-5.6. Arena's frontend coding leaderboard, which tracks nearly 470,000 votes across 96 models, uses a blind comparison method: voters see two models without knowing which is which.
A Clean Sweep of Categories
Kimi-K3 won six of the seven coding categories tested. Brand & Marketing, Reference-Based Design, Data & Analytics — each one went to Moonshot's model. Claude Fable 5 managed to hold on to only one category: Gaming. Anthropic still has nine models in the overall top 20, but its flagship lost its crown.
Pricing and Open-Source Plans
Moonshot is pricing Kimi-K3 aggressively: $3 per million input tokens and $15 per million output tokens. That's roughly a third of Claude Fable 5's $10 and $50 rates. The company also announced it will release Kimi-K3's full model weights by July 27, 2026, letting anyone run the leading coding model for free. That move could reshape how developers choose their tools.
Alibaba's Claude Code Ban
Meanwhile, Alibaba told its staff to stop using Claude Code starting July 10, 2026, citing security concerns. The directive comes as Chinese tech companies increasingly rely on homegrown models. Mark Zuckerberg's Muse Spark 1.1 sits at 11th on the leaderboard, days after its rollout.
All eyes are now on the open-source release date: July 27, 2026. Until then, Kimi-K3 remains a proprietary model — but one that's already changed the frontend coding race.
