GPT-5.4 vs Kimi K2.6
Wins 6 of 8 areas
Coding · Agents · Reasoning · Facts · Math · Long documents
Wins 1 of 8 areas
Images and charts
GPT-5.4 is the stronger all-rounder.Kimi K2.6 is cheaper and better at images and charts.
Scores updated · 78 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 math70GPT-5.47 of 7 tests
- AgentsCarrying out multi-step tasks on its own51GPT-5.45 of 7 tests · 1 tie
- ReasoningHard problems that need careful thinking30GPT-5.43 of 4 tests · 1 tie
- FactsGetting facts right instead of making them up31GPT-5.43 of 4 tests
- CodingWriting and fixing software43GPT-5.44 of 8 tests · 1 tie
- Long documentsFinding answers in very long texts10GPT-5.41 of 1 test
- Images and chartsUnderstanding pictures, charts and video12Kimi K2.62 of 3 tests
- Following instructionsDoing exactly what it is asked11Even1 each
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 72% 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-5.4 pulls ahead
- Proof problems from the US Math OlympiadUSAMO 2026+44points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+23.4points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+21.4points ahead
Where Kimi K2.6 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+51.2points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+5points ahead
- Reasoning about charts from research papersCharXiv (RQ)+3.9points ahead
Every test, side by side
All 78 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.4
- Terminal-Bench HardGPT-5.4 by 13.757.643.9+13.7
- Terminal-Bench 2.1GPT-5.4 by 12.478.365.9+12.4
- LiveBench · Agentic CodingGPT-5.4 by 6.953.846.9+6.9
- SciCodeGPT-5.4 by 5.156.651.5+5.1
- SWE-bench MultilingualKimi K2.6 by 571.776.7+5
- LMArena · WebDevKimi K2.6 by 44 rating points14651509+44 rating
- LiveBench · CodingKimi K2.6 by 1.177.578.6+1.1
- SWE-bench Protie57.758.6tie
AgentsGPT-5.4
- τ-Bench V3 · BankingGPT-5.4 by 16.339.623.3+16.3
- GDPValGPT-5.4 by 10.437.427+10.4
- AA ApexAgentsGPT-5.4 by 4.833.328.5+4.8
- BrowseCompKimi K2.6 by 3.682.786.3+3.6
- AA IT-Bench SREGPT-5.4 by 3.334.531.2+3.3
- OSWorld-VerifiedGPT-5.4 by 1.97573.1+1.9
- MCP Atlastie67.268.1tie
ReasoningGPT-5.4
- CritPtGPT-5.4 by 15.423.48+15.4
- LiveBench · ReasoningGPT-5.4 by 8.788.179.4+8.7
- Humanity's Last ExamGPT-5.4 by 6.243.737.5+6.2
- GPQA Diamondtie9291.1tie
FactsGPT-5.4
- AA-Omniscience · Non-hallucinationKimi K2.6 by 51.28.359.5+51.2
- AA-Omniscience · AccuracyGPT-5.4 by 18.350.932.6+18.3
- SimpleQA VerifiedGPT-5.4 by 10.245.134.9+10.2
- Vectara HHEM hallucination ratelower is betterGPT-5.4 by 3.8710.8+3.8
Images and chartsKimi K2.6
- CharXiv (RQ)Kimi K2.6 by 3.982.886.7+3.9
- LMArena · VisionGPT-5.4 by 20 rating points13021282+20 rating
- MMMU-ProKimi K2.6 by 178.479.4+1
MathGPT-5.4
- USAMO 2026GPT-5.4 by 4495.251.2+44
- FrontierMath Tier 4GPT-5.4 by 23.44925.6+23.4
- FrontierMath Tiers 1-3 (v2)GPT-5.4 by 21.478.657.2+21.4
- LiveBench · MathematicsGPT-5.4 by 9.994.284.3+9.9
- IMOAnswerBenchGPT-5.4 by 5.491.486+5.4
- HMMT Feb. 2026GPT-5.4 by 597.792.7+5
- AIME 2026GPT-5.4 by 2.899.296.4+2.8
Following instructionsEven
- LiveBench · Instruction FollowingGPT-5.4 by 5.870.264.4+5.8
- IFBenchKimi K2.6 by 2.173.976+2.1
Other results42 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.
- Apex (Pass@1)GPT-5.4 by 30.154.124+30.1
- AA Agentic IndexGPT-5.4 by 22.144.222.1+22.1
- DeepSearchQA (accuracy)Kimi K2.6 by 19.363.783+19.3
- GDPval-AA (Elo)GPT-5.4 by 192 rating points16741482+192 rating
- BabyVisionKimi K2.6 by 18.849.768.5+18.8
- livebench_data_analysisGPT-5.4 by 14.279.365.1+14.2
- frontiermath_tier_4_v1GPT-5.4 by 12.527.114.6+12.5
- AA IntelligenceGPT-5.4 by 123927+12
- baby_vision_with_pythonGPT-5.4 by 11.780.268.5+11.7
- AgentWorldBench - SearchGPT-5.4 by 9.837.327.5+9.8
- SWE-QAGPT-5.4 by 9.781.371.6+9.7
- Artificial Analysis Coding IndexGPT-5.4 by 9.371.161.8+9.3
- HiL-Bench (Tools-allowed)Kimi K2.6 by 99.718.7+9
- τ²-Bench Telecom (AA run)Kimi K2.6 by 8.887.195.9+8.8
- Terminal-Bench 2.0GPT-5.4 by 8.475.166.7+8.4
- LiveBenchGPT-5.4 by 8.180.372.2+8.1
- AgentWorldBench - OSGPT-5.4 by 7.868.660.8+7.8
- AgentWorldBench - SWEGPT-5.4 by 7.566.358.8+7.5
- livebench_languageGPT-5.4 by 7.582.675.1+7.5
- MCPMarkGPT-5.4 by 6.662.555.9+6.6
- AgentWorldBench - MCPGPT-5.4 by 4.970.165.2+4.9
- AgentWorldBench - OverallGPT-5.4 by 4.958.353.4+4.9
- ToolathlonGPT-5.4 by 4.654.650+4.6
- DeepSearchQA (F1)Kimi K2.6 by 4.478.683+4.4
- vectara_factual_consistencyGPT-5.4 by 3.89389.2+3.8
- BrowseComp (Agent Swarm)Kimi K2.6 by 3.682.786.3+3.6
- vectara_avg_summary_lengthKimi K2.6 by 35 rating points81.7116.7+35 rating
- charxiv_rq_with_pythonGPT-5.4 by 3.39086.7+3.3
- mathvision_with_pythonGPT-5.4 by 2.996.193.2+2.9
- Apex Shortlist (Pass@1)GPT-5.4 by 2.678.175.5+2.6
- mmmu_pro_with_pythonGPT-5.4 by 282.180.1+2
- HLE (with tools)Kimi K2.6 by 1.952.154+1.9
- AgentWorldBench - WebGPT-5.4 by 1.651.850.2+1.6
- V* (w/ python)GPT-5.4 by 1.598.496.9+1.5
- AgentWorldBench - TerminalGPT-5.4 by 1.253.752.5+1.2
- MATH-VisionKimi K2.6 by 1.29293.2+1.2
- AgentWorldBench - AndroidGPT-5.4 by 1.16058.9+1.1
- Chinese SimpleQA (C-SimpleQA)tie76.875.9tie
- MCPAtlas Public (Pass@1)tie67.266.6tie
- AA-Omnisciencetie5.85.3tie
- MMLU-Protie87.587.1tie
- vectara_answer_ratetie99.999.7tie
Questions people ask
Which is better, GPT-5.4 or Kimi K2.6?
GPT-5.4 wins six of the eight areas we test: coding, agents, reasoning, facts, math and long documents. Kimi K2.6 wins images and charts, and costs 72% less. They are level on following instructions.
Which is better for coding?
GPT-5.4. It wins 4 of the 8 coding tests both models report; Kimi K2.6 wins 3, and 1 is a tie.
Which is cheaper?
GPT-5.4 costs $2.50 per million input tokens and $15.00 per million output tokens; Kimi K2.6 costs $0.95 and $4.00. That makes Kimi K2.6 about 72% cheaper for the same work.
How do you compare the two?
We use the 78 benchmark tests both models have published scores on. The verdict counts the 36 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 42 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.