GPT-5.5 vs Kimi K3
Wins 3 of 8 areas
Reasoning · Facts · Math
Wins 3 of 8 areas
Coding · Agents · Long documents
The two are evenly matched.GPT-5.5 is better at math and reasoning; Kimi K3 at agents and coding.
Both rank among the ten best models we track in agents and long documents.
Scores updated · 59 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.
- AgentsCarrying out multi-step tasks on its own17Kimi K37 of 8 tests
- CodingWriting and fixing software03Kimi K33 of 5 tests · 2 ties
- Long documentsFinding answers in very long texts01Kimi K31 of 1 test
- MathCompetition and research-level math50GPT-5.55 of 5 tests
- ReasoningHard problems that need careful thinking32GPT-5.53 of 6 tests · 1 tie
- FactsGetting facts right instead of making them up21GPT-5.52 of 3 tests
- Images and chartsUnderstanding pictures, charts and video00Even0 each · 1 tie
- Following instructionsDoing exactly what it is asked00Even0 each · 1 tie
Images and charts, long documents and following instructions rest on a single test each.
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 K3 costs 49% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Kimi K3 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+35.8points ahead
- Real work tasks from 44 professionsGDPVal+9points ahead
- Multi-step tasks using many real software toolsMCP Atlas+8.9points ahead
Where GPT-5.5 pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+33.5points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+24.6points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+13.1points ahead
Every test, side by side
All 59 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingKimi K3
- LMArena · WebDevKimi K3 by 142 rating points15131655+142 rating
- LiveBench · Agentic CodingKimi K3 by 8.25462.2+8.2
- SciCodeKimi K3 by 3.755.859.5+3.7
- LiveBench · Codingtie82.281.5tie
- Terminal-Bench 2.1tie84.385tie
AgentsKimi K3
- GDPValKimi K3 by 942.751.7+9
- MCP AtlasKimi K3 by 8.975.384.2+8.9
- τ-Bench V3 · BankingKimi K3 by 73946+7
- BrowseCompKimi K3 by 6.884.491.2+6.8
- OSWorld-VerifiedKimi K3 by 6.178.784.8+6.1
- AA ApexAgentsKimi K3 by 3.637.741.3+3.6
- Terminal-Bench 4.0GPT-5.5 by 214.612.6+2
- AA IT-Bench SREKimi K3 by 1.945.847.7+1.9
ReasoningGPT-5.5
- ARC-AGI-2GPT-5.5 by 24.68560.4+24.6
- SimpleBenchGPT-5.5 by 8.36960.7+8.3
- CritPtGPT-5.5 by 3.727.123.4+3.7
- Humanity's Last ExamKimi K3 by 1.145.846.9+1.1
- LiveBench · ReasoningKimi K3 by 189.790.7+1
- GPQA Diamondtie93.593.5tie
FactsGPT-5.5
- AA-Omniscience · Non-hallucinationKimi K3 by 35.81146.8+35.8
- SimpleQA VerifiedGPT-5.5 by 12.46350.6+12.4
- AA-Omniscience · AccuracyGPT-5.5 by 10.45847.6+10.4
MathGPT-5.5
- FrontierMath Tier 4GPT-5.5 by 33.572.539+33.5
- FrontierMath Tiers 1-3 (v2)GPT-5.5 by 13.185.372.2+13.1
- LiveBench · MathematicsGPT-5.5 by 11.595.984.4+11.5
- AIME 2026GPT-5.5 by 3.310096.7+3.3
- HMMT Feb. 2026GPT-5.5 by 1.598.597+1.5
Following instructionsEven
- LiveBench · Instruction Followingtie70.771.4tie
Other results29 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.
- SWE-MarathonKimi K3 by 301242+30
- BabyVisionKimi K3 by 29.855.985.7+29.8
- GDPval-AA v2 EloKimi K3 by 195 rating points14911686+195 rating
- ToolathlonKimi K3 by 17.655.673.2+17.6
- GDPval-AA v2Kimi K3 by 173 rating points15091682+173 rating
- WorldVQAKimi K3 by 16.434.651+16.4
- JobBenchKimi K3 by 14.638.352.9+14.6
- AA Agentic IndexKimi K3 by 13.337.350.6+13.3
- MLS Bench Litelower is betterGPT-5.5 by 12.835.548.3+12.8
- ResearchRubricsKimi K3 by 12.26476.2+12.2
- OfficeQA ProKimi K3 by 9.254.163.3+9.2
- PostTrainBenchKimi K3 by 8.228.436.6+8.2
- HLE (with tools)Kimi K3 by 7.652.259.8+7.6
- NL2RepoKimi K3 by 7.350.758+7.3
- Program BenchKimi K3 by 770.877.8+7
- SpreadsheetBench 2Kimi K3 by 5.729.134.8+5.7
- AA IntelligenceKimi K3 by 5.238.443.6+5.2
- matharena_visual_math_overallGPT-5.5 by 3.694.991.3+3.6
- Toolathlon VerifiedKimi K3 by 373.576.5+3
- livebench_data_analysisGPT-5.5 by 2.981.678.7+2.9
- DeepSWEGPT-5.5 by 2.57067.5+2.5
- DeepSWE 1.1Kimi K3 by 26769+2
- livebench_languageGPT-5.5 by 1.987.485.5+1.9
- CyberGymGPT-5.5 by 1.881.880+1.8
- Agents' Last ExamKimi K3 by 1.726.628.3+1.7
- MCP Mark VerifiedKimi K3 by 1.692.994.5+1.6
- Artificial Analysis Coding IndexKimi K3 by 1.374.976.2+1.3
- AA-Omnisciencetie20.519.7tie
- ARC-AGI-1tie9594.5tie
Questions people ask
Which is better, GPT-5.5 or Kimi K3?
GPT-5.5 and Kimi K3 each win three of the eight areas we test. GPT-5.5 wins reasoning, facts and math; Kimi K3 wins coding, agents and long documents. Kimi K3 costs 49% less. They are level on images and charts and following instructions.
Which is better for coding?
Kimi K3. It wins 3 of the 5 coding tests both models report; GPT-5.5 wins none, and 2 are ties.
Which is cheaper?
GPT-5.5 costs $5.00 per million input tokens and $30.00 per million output tokens; Kimi K3 costs $3.00 and $15.00. That makes Kimi K3 about 49% cheaper for the same work.
How do you compare the two?
We use the 59 benchmark tests both models have published scores on. The verdict counts the 30 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 29 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.