Arena says Kimi K3.1 Pro (Max) entered its Agent Arena at #11, which would make it the top open model there once weights ship
Kimi K3.1 Pro (Max) by @Kimi_Moonshot just landed in the Agent Arena at #11 with +7.6% net improvement across +4K real-world agentic sessions from our global community of users. Kimi K3.1 Pro (Max) would be the new #1 open model in the Agent Arena overtaking DeepSeek-v4.1-Flash! Open weights coming soon, stay tuned. Kimi K3.1 Pro (Max) is a +6 percentage pt improvement over Kimi K3 (Max) at +1.6% (#25 -> #11). And with a $1.03 median cost per task in Agent Arena, Kimi K3.1 Pro is 13% less than its previous variant. See its placement just off the Pareto frontier in thread. Across Agent Arena categories, Kimi K3.1 Pro (Max) ranks #8 in Work, #9 in Code, and #10 in Chat. By signal, it’s the strongest in Praise vs. Complaint (implicit sentiment in user reactions) with +19% at #8. Across other signals it landed: - #11 in Steerability (+8.3%) - #15 in Bash Reliability (+8.3%) - #26 in Confirmed Success (+2%) Huge congrats to the @Kimi_Moonshot team on this release!
Context
Agent Arena ranks models by how real users rate them on agent tasks, so it is an outside check on Moonshot's own launch claims. A top open model this close to closed leaders would give companies a cheaper agent option they can run themselves.
Moonshot launched Kimi K3.1 Pro on October 23 and promised open weights by November 4. Arena is the company behind the LMArena leaderboards, and its Agent Arena scores models on real agent sessions from its users. The post was published at 10:57 AM Eastern on October 24.
Deep dive
Arena posted at 10:57 AM Eastern on October 24 that Kimi K3.1 Pro (Max) from Moonshot landed at #11 in its Agent Arena, with a 7.6% net improvement over baseline across more than 4,000 real agent sessions from its users. Arena said that once the weights are released it would be the new #1 open model on the board, ahead of DeepSeek V4.1 Flash.
The jump from the previous model is large by this measure. Arena says Kimi K3 (Max) sat at #25 with 1.6%, so the new version gained about six percentage points. It reports a median cost of $1.03 per task, 13% less than the earlier variant, which places it just off the line of best cost for performance. By category it ranks #8 in Work, #9 in Code and #10 in Chat.
Arena also breaks results down by user signals. Kimi K3.1 Pro did best on praise versus complaint, the implied sentiment in user reactions, at #8. It placed #11 for steerability, #15 for reliability in shell commands and #26 for confirmed task success, the weakest of its signals.
The chart shows the top ten are all closed models from Anthropic, OpenAI, Google and Meta, led by Claude Opus 5.5 (Max) at 20.8%, so the gap between the best open model and the frontier remains wide on this board.
What to expect next
The ranking may move as more sessions come in, since Arena's error bars on new entries are wide. The bigger test comes when Moonshot publishes the weights by November 4, which would let others host the model and confirm its cost and quality outside Moonshot's own API. Watch whether DeepSeek, Qwen or Meta answer with open releases that retake the top open spot.