‹ PublicAI Index
The LLM benchmark aggregator.
Kimi K2 Thinking
Moonshot · 1T · open weights
Strongest in Legal (#43 of 173), weakest in Harm refusal (#185 of 292). Above par in 14 of 20 scopes. Among the models it meets almost everywhere, it finishes behind GPT-5 and Claude Fable 5 and ahead of O4 Mini and GPT-4.1.
§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Professional53.7−8.7#50/1941/2
Legal55.6−10.3#43/1731/1
Medical54.8−11.7#43/1491/1
Finance52.3−10.1#82/1621/1
Core abilities54−11.3#58/1991/4
General intelligence55.7−13.7#47/1781/3
Reasoning50.4−15#95/1952/5
Mathematics55.1−8.7#53/1521/2
Science51.7−11.8#72/1221/1
Reasoning44.5−24.5#92/1471/3
Knowledge49.9−12.6#115/1971/3
Academic knowledge49.8−14.1#80/1231/1
Safety52.2−9.2#125/3291/3
Fairness56.7−13.3#53/2921/2
Secure code56.6−9.6#99/2701/1
Toxicity avoidance54.8−4.2#110/2681/1
Jailbreak resistance51.4−13.6#164/2681/1
Harm refusal48−13.5#185/2921/2
Coding45.2−21#126/1791/6
Agentic coding44.8−22.1#122/1731/5
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Kimi K2 Thinking, left for the other.
§ 3 · Sources
Where the numbers come from
4 publications, 17 figures. Every one links to the page it was read from.
Kagi LLM Benchmark 64.4%
SimpleBench 39.6%
Vals · CaseLaw 65.7%Vals · LegalBench 80.2%Vals · CorpFin 60.57%Vals · TaxEval 71.71%Vals · MedQA 92.59%Vals · SWE-bench Verified 60.2% (Mini-SWE-agent)Vals · GPQA Diamond 78.54%Vals · MMLU Pro 81.07%Vals · AIME 85.42%
Enkrypt · Jailbreak risk 11.6%Enkrypt · Harmful content risk 26.1%Enkrypt · CBRN risk 24.3%Enkrypt · Toxicity risk 2.9%Enkrypt · Bias risk 67.4%Enkrypt · Insecure code risk 19.6%
Badge
[](https://publicai.io/model-index/m/kimi-k2-thinking)