Event · Hands-on reviews

Kimi K3.1 Pro: what testers say, from "can't tell it from Opus" to "limit gone in one minute"

Moonshot AI's open-weight Kimi K3.1 Pro landed Oct 24, and X filled with hands-on results. Leaving out company and staff accounts, here are tests with numbers or outputs, good and bad.

1. Negative@PawelHuryn · X ↗

   105 planted bugs: Opus 41.7, Kimi 36

   Pawel Huryn planted 105 bugs across two repos and had models find and fix them. DeepSeek V4.1 Flash scored 42.7, Muse Wonder Full 42, Claude Opus 5.5 41.7 and Kimi K3.1 Pro 36, ahead of Muse Spark 1.3 (32.2). "I'm sorry," he wrote.

2. Mixed@Da7_Tech · X ↗

   "Best open model, practically unusable on its own subscription"

   Da7 Tech called it "the best open model in the world right now. No contest," but said a single ship-building test on the $39 plan ate 11.6% of the monthly quota.

3. Negative@MiaAI_lab · X ↗

   "My 5-hour limit was done after one minute, and nothing was built"

   MiaAI_lab called the usage limits "absolutely disgusting and shameful," saying the 5-hour limit ran out a minute into the prompt with nothing built, and that it's worse than before: "the WORST subscription of them all." The post drew 922 likes.

4. Mixed@yohakujpn · X ↗

   Black hole build: Kimi 30 minutes, Opus 48

   Yohaku had both models build a black hole and posted the results side by side: 30 minutes for Kimi K3.1, 48 for Opus 5.5, then asked which did better. 393 likes.

5. Positive@davidliyangbing · X ↗

   40-minute helicopter demo: "can't tell the difference from Opus 5.5"

   David Liyangbing let it run on a helicopter demo for 40 minutes: "I genuinely can't tell the difference between this and Opus 5.5 anymore."

6. Positive@AI_Screening · X ↗

   Single-file three.js test: "Kimi cooked Astra"

   AI_Screening ran Kimi K3.1 Pro, Opus 5.5 and GPT-6 Astra on a single-HTML three.js scene: "Kimi K3.1 cooked Astra… now competing at the frontier." Over 1,300 likes.

7. Mixed@LLMJunky · X ↗

   Rocket League Bench: best open model, a generation behind on 3D

   LLMJunky called it "definitely the best open model that I've tested so far," but "a full generation behind the frontier on 3D modeling," behind Opus 5, GPT-6 Astra, Fable 5.1 and Opus 5.5 on that one test.

8. Mixed@ml_angelopoulos · X ↗

   "Similar to Opus 5, so China lags the frontier by about 3 months"

   Anastasios Angelopoulos called it the #1 open-source model but said it places similarly to Opus 5, released three months ago, so "China is lagging the frontier by about 3 months." Kimi K3 was #1 on Code Arena at launch; K3.1 is close but "substantially off," he said.