GPT-5.2 vs Kimi K2.6
Wins 4 of 8 areas
Reasoning · Facts · Math · Long documents
Wins 4 of 8 areas
Coding · Agents · Images and charts · Following instructions
The two are evenly matched.GPT-5.2 is better at math and reasoning; Kimi K2.6 at agents and images and charts.
Scores updated · 54 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- MathCompetition and research-level math50GPT-5.25 of 6 tests · 1 tie
- ReasoningHard problems that need careful thinking20GPT-5.22 of 4 tests · 2 ties
- FactsGetting facts right instead of making them up21GPT-5.22 of 4 tests · 1 tie
- Long documentsFinding answers in very long texts10GPT-5.21 of 1 test
- AgentsCarrying out multi-step tasks on its own14Kimi K2.64 of 5 tests
- Images and chartsUnderstanding pictures, charts and video03Kimi K2.63 of 3 tests
- CodingWriting and fixing software24Kimi K2.64 of 8 tests · 2 ties
- Following instructionsDoing exactly what it is asked01Kimi K2.61 of 2 tests · 1 tie
Long documents rests on a single test.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
Kimi K2.6 costs 69% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where GPT-5.2 pulls ahead
- Real work tasks from 44 professionsGDPVal+21.3points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+11.7points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+10.2points ahead
Where Kimi K2.6 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+40.7points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+28.7points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+20.5points ahead
Every test, side by side
All 54 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingKimi K2.6
- SWE-bench ProKimi K2.6 by 28.729.958.6+28.7
- SWE-bench MultilingualKimi K2.6 by 1066.776.7+10
- LMArena · WebDevKimi K2.6 by 93 rating points14161509+93 rating
- LiveBench · Agentic CodingGPT-5.2 by 3.450.346.9+3.4
- Terminal-Bench HardGPT-5.2 by 3.14743.9+3.1
- LiveBench · CodingKimi K2.6 by 2.576.178.6+2.5
- SciCodetie52.151.5tie
- SWE-bench Verifiedtie8080.2tie
AgentsKimi K2.6
- GDPValGPT-5.2 by 21.348.327+21.3
- BrowseCompKimi K2.6 by 20.565.886.3+20.5
- τ-Bench V3 · BankingKimi K2.6 by 12.211.123.3+12.2
- MCP AtlasKimi K2.6 by 7.560.668.1+7.5
- AA ApexAgentsKimi K2.6 by 5.52328.5+5.5
ReasoningGPT-5.2
- LiveBench · ReasoningGPT-5.2 by 3.883.279.4+3.8
- CritPtGPT-5.2 by 3.611.68+3.6
- GPQA Diamondtie90.391.1tie
- Humanity's Last Examtie37.737.5tie
FactsGPT-5.2
- AA-Omniscience · Non-hallucinationKimi K2.6 by 40.718.859.5+40.7
- AA-Omniscience · AccuracyGPT-5.2 by 11.744.332.6+11.7
- SimpleQA VerifiedGPT-5.2 by 2.237.134.9+2.2
- Vectara HHEM hallucination ratelower is bettertie10.810.8tie
Images and chartsKimi K2.6
- MMMU-ProKimi K2.6 by 4.874.679.4+4.8
- CharXiv (RQ)Kimi K2.6 by 4.682.186.7+4.6
- LMArena · VisionKimi K2.6 by 36 rating points12461282+36 rating
MathGPT-5.2
- FrontierMath Tiers 1-3 (v2)GPT-5.2 by 10.267.457.2+10.2
- LiveBench · MathematicsGPT-5.2 by 8.993.284.3+8.9
- FrontierMath Tier 4GPT-5.2 by 6.131.725.6+6.1
- HMMT Feb. 2026GPT-5.2 by 4.39792.7+4.3
- AIME 2026GPT-5.2 by 1.998.396.4+1.9
- IMOAnswerBenchtie86.386tie
Following instructionsKimi K2.6
- LiveBench · Instruction FollowingKimi K2.6 by 2.661.864.4+2.6
- IFBenchtie75.476tie
Other results21 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- AA Agentic IndexGPT-5.2 by 38.160.222.1+38.1
- Artificial Analysis Coding IndexKimi K2.6 by 13.148.761.8+13.1
- livebench_data_analysisGPT-5.2 by 13.178.265.1+13.1
- Terminal-Bench 2.0Kimi K2.6 by 12.75466.7+12.7
- DeepSearchQA (F1)Kimi K2.6 by 11.771.383+11.7
- τ²-Bench Telecom (AA run)Kimi K2.6 by 11.184.895.9+11.1
- MATH-VisionKimi K2.6 by 10.28393.2+10.2
- HLE (with tools)Kimi K2.6 by 8.545.554+8.5
- vectara_avg_summary_lengthGPT-5.2 by 69.6 rating points186.3116.7+69.6 rating
- AA-OmniscienceKimi K2.6 by 6.2-0.95.3+6.2
- livebench_languageGPT-5.2 by 4.779.875.1+4.7
- frontiermath_tier_4_v1GPT-5.2 by 4.218.814.6+4.2
- ToolathlonKimi K2.6 by 3.746.350+3.7
- AA IntelligenceGPT-5.2 by 3.430.427+3.4
- LiveBenchGPT-5.2 by 2.674.872.2+2.6
- GDPval-AA (Elo)Kimi K2.6 by 20 rating points14621482+20 rating
- GPQA (unspecified)GPT-5.2 by 1.492.491+1.4
- LiveCodeBenchtie8989.6tie
- MMLU-Protie86.787.1tie
- vectara_answer_ratetie10099.7tie
- vectara_factual_consistencytie89.289.2tie
Questions people ask
Which is better, GPT-5.2 or Kimi K2.6?
GPT-5.2 and Kimi K2.6 each win four of the eight areas we test. GPT-5.2 wins reasoning, facts, math and long documents; Kimi K2.6 wins coding, agents, images and charts and following instructions. Kimi K2.6 costs 69% less.
Which is better for coding?
Kimi K2.6. It wins 4 of the 8 coding tests both models report; GPT-5.2 wins 2, and 2 are ties.
Which is cheaper?
GPT-5.2 costs $1.75 per million input tokens and $14.00 per million output tokens; Kimi K2.6 costs $0.95 and $4.00. That makes Kimi K2.6 about 69% cheaper for the same work.
How do you compare the two?
We use the 54 benchmark tests both models have published scores on. The verdict counts the 33 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 21 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.