GPT-5.2 vs Kimi K2.5
Wins 4 of 8 areas
Reasoning · Facts · Math · Following instructions
Wins 2 of 8 areas
Agents · Images and charts
GPT-5.2 wins more areas, narrowly.Kimi K2.5 is cheaper and better at images and charts.
Scores updated · 66 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.
- ReasoningHard problems that need careful thinking40GPT-5.24 of 4 tests
- MathCompetition and research-level math30GPT-5.23 of 3 tests
- FactsGetting facts right instead of making them up31GPT-5.23 of 4 tests
- Following instructionsDoing exactly what it is asked10GPT-5.21 of 1 test
- Images and chartsUnderstanding pictures, charts and video14Kimi K2.54 of 6 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own23Kimi K2.53 of 5 tests
- CodingWriting and fixing software33Even3 each
- Long documentsFinding answers in very long texts11Even1 each
Following instructions 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 K2.5 costs 77% 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
- Abstract visual puzzles that people can solveARC-AGI-2+41.1points ahead
- Real work tasks from 44 professionsGDPVal+31.1points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+12.2points ahead
Where Kimi K2.5 pulls ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+20.8points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+15.5points ahead
- Reads text in imagesOCRBench+11.6points ahead
Every test, side by side
All 66 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingEven
- SWE-bench ProKimi K2.5 by 20.829.950.7+20.8
- Terminal-Bench HardGPT-5.2 by 12.24734.8+12.2
- SWE-bench MultilingualKimi K2.5 by 6.366.773+6.3
- SWE-bench VerifiedGPT-5.2 by 3.28076.8+3.2
- SciCodeGPT-5.2 by 3.152.149+3.1
- LMArena · WebDevKimi K2.5 by 20 rating points14161436+20 rating
AgentsKimi K2.5
- GDPValGPT-5.2 by 31.148.317.2+31.1
- AA ApexAgentsGPT-5.2 by 11.52311.5+11.5
- BrowseCompKimi K2.5 by 9.165.874.9+9.1
- MCP AtlasKimi K2.5 by 3.260.663.8+3.2
- τ-Bench V3 · BankingKimi K2.5 by 3.111.114.2+3.1
ReasoningGPT-5.2
- ARC-AGI-2GPT-5.2 by 41.152.911.8+41.1
- CritPtGPT-5.2 by 8.511.63.1+8.5
- Humanity's Last ExamGPT-5.2 by 737.730.7+7
- GPQA DiamondGPT-5.2 by 2.490.387.9+2.4
FactsGPT-5.2
- AA-Omniscience · Non-hallucinationKimi K2.5 by 15.518.834.3+15.5
- AA-Omniscience · AccuracyGPT-5.2 by 9.144.335.2+9.1
- Vectara HHEM hallucination ratelower is betterGPT-5.2 by 3.410.814.2+3.4
- SimpleQA VerifiedGPT-5.2 by 2.837.134.3+2.8
Images and chartsKimi K2.5
- OCRBenchKimi K2.5 by 11.680.792.3+11.6
- MathVistaKimi K2.5 by 7.382.890.1+7.3
- CharXiv (RQ)GPT-5.2 by 4.682.177.5+4.6
- LMArena · VisionKimi K2.5 by 23 rating points12461269+23 rating
- Video-MMEKimi K2.5 by 1.48687.4+1.4
- MMMU-Protie74.675.4tie
MathGPT-5.2
- HMMT Feb. 2026GPT-5.2 by 9.99787.1+9.9
- IMOAnswerBenchGPT-5.2 by 4.586.381.8+4.5
- AIME 2026GPT-5.2 by 2.598.395.8+2.5
Long documentsEven
- LongBench v2Kimi K2.5 by 6.554.561+6.5
- AA-LCRGPT-5.2 by 4.782.778+4.7
Other results35 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 38.560.221.7+38.5
- ARC-AGI-1GPT-5.2 by 20.986.265.3+20.9
- ToolathlonGPT-5.2 by 18.546.327.8+18.5
- WorldVQAKimi K2.5 by 18.32846.3+18.3
- SimpleVQAKimi K2.5 by 15.455.871.2+15.4
- frontiermath_tier_4_v1GPT-5.2 by 14.618.84.2+14.6
- τ²-Bench Telecom (AA run)Kimi K2.5 by 11.184.895.9+11.1
- InfoVQA (val)Kimi K2.5 by 8.68492.6+8.6
- vectara_answer_rateGPT-5.2 by 7.810092.2+7.8
- vectara_avg_summary_lengthGPT-5.2 by 74.3 rating points186.3112+74.3 rating
- AA IntelligenceGPT-5.2 by 6.930.423.5+6.9
- HMMT Nov. 2025GPT-5.2 by 6.695.889.2+6.6
- AA-OmniscienceGPT-5.2 by 6.4-0.9-7.3+6.4
- matharena_visual_math_overallGPT-5.2 by 5.986.580.6+5.9
- DeepSearchQA (F1)Kimi K2.5 by 5.871.377.1+5.8
- LiveBenchGPT-5.2 by 5.774.869.1+5.7
- MotionBenchKimi K2.5 by 5.664.870.4+5.6
- HLE (with tools)Kimi K2.5 by 4.745.550.2+4.7
- HMMT Feb. 2025GPT-5.2 by 4.610095.4+4.6
- HMMT 2025GPT-5.2 by 499.495.4+4
- ZeroBench (w/ tools)Kimi K2.5 by 4711+4
- AIME 2025GPT-5.2 by 3.910096.1+3.9
- vectara_factual_consistencyGPT-5.2 by 3.489.285.8+3.4
- LongVideoBenchKimi K2.5 by 3.376.579.8+3.3
- Terminal-Bench 2.0GPT-5.2 by 3.25450.8+3.2
- OmniDocBench 1.5Kimi K2.5 by 3.185.788.8+3.1
- ZeroBenchKimi K2.5 by 2911+2
- Artificial Analysis Coding IndexGPT-5.2 by 1.948.746.8+1.9
- swe_bench_bashKimi K2.5 by 1.86970.8+1.8
- MATH-VisionKimi K2.5 by 1.28384.2+1.2
- VideoMMMUtie85.986.6tie
- MMLU-Protie86.787.1tie
- MMVUtie80.880.4tie
- PaperBenchtie63.763.5tie
- τ²-Benchtie85.585.4tie
Questions people ask
Which is better, GPT-5.2 or Kimi K2.5?
GPT-5.2 wins four of the eight areas we test: reasoning, facts, math and following instructions. Kimi K2.5 wins agents and images and charts, and costs 77% less. They are level on coding and long documents.
Which is better for coding?
Neither. They win 3 coding tests each of the 6 both models report.
Which is cheaper?
GPT-5.2 costs $1.75 per million input tokens and $14.00 per million output tokens; Kimi K2.5 costs $0.60 and $3.00. That makes Kimi K2.5 about 77% cheaper for the same work.
How do you compare the two?
We use the 66 benchmark tests both models have published scores on. The verdict counts the 31 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 35 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.