← Back to New
SCORE RESULT

Kimi K3 enters ClawProBench at #1

A day after release, independent boards post their first Kimi K3 results: first place among 64 models on ClawProBench and a close second of 32 on LiveBench Reasoning.

Event date Published

First place on ClawProBench, a point ahead of the standing leader

Kimi K3's first ClawProBench result is 78.4, first among the 64 models on the board and 1.2 points ahead of GLM-5.2 at 77.2. GLM-5.2's score did not change: Kimi K3 entered above the standing leader rather than displacing a declining one. On LiveBench Reasoning it debuted at 90.67, second of 32 models, 0.98 behind GPT-5.6 Sol and just 0.04 ahead of GPT-5.6 Terra — effectively a tie for second with Terra.

ClawProBench
78.4 #1 of 64
Ahead of GLM-5.2 at 77.2 and Agnes 2.0 Flash at 74.6.
LiveBench Reasoning
90.67 #2 of 32
Behind GPT-5.6 Sol at 91.65; effectively tied with GPT-5.6 Terra at 90.63.

Kimi K3's entry against each board's leaders

Two panels. On ClawProBench, Kimi K3 debuts first of 64 models at 78.4, ahead of GLM-5.2 at 77.2, Agnes 2.0 Flash at 74.6, and InternLM S2 Preview at 73.7. On LiveBench Reasoning, Kimi K3 debuts second of 32 at 90.67, behind GPT-5.6 Sol at 91.65 and just ahead of GPT-5.6 Terra at 90.63.
PanelItemBenchmark score
ClawProBench — Kimi K3 #1 of 64Kimi K378.4
ClawProBench — Kimi K3 #1 of 64GLM-5.277.2
ClawProBench — Kimi K3 #1 of 64Agnes 2.0 Flash74.6
ClawProBench — Kimi K3 #1 of 64InternLM S2 Preview73.7
LiveBench Reasoning — Kimi K3 #2 of 32GPT-5.6 Sol91.65
LiveBench Reasoning — Kimi K3 #2 of 32Kimi K390.67
LiveBench Reasoning — Kimi K3 #2 of 32GPT-5.6 Terra90.63

Kimi K3 is highlighted. Each panel uses a labeled local scale that does not start at 0, so nearby gaps stay readable; compare positions only within the same panel.

The rest of the first wave is strong but uneven

These results land a day after the July 16 launch coverage, on boards that coverage did not include, and the wider wave is mixed: sixth of 32 on LiveBench overall at 78.5, fourth of 16 on DeepSWE, fourth of 43 on Vals ProofBench, and twenty-third of 32 on LiveBench Mathematics. Leading one board does not make an overall winner; the spread shows where the debut is strongest.