GPT-5 vs Kimi K2 Instruct
Wins 7 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Long documents · Following instructions
Wins 0 of 7 areas
—
GPT-5 is the stronger all-rounder.Kimi K2 Instruct is cheaper.
Scores updated · 64 tests both models report · How we compare
Where each one wins
Tests won in each of the seven areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- CodingWriting and fixing software60GPT-56 of 6 tests
- ReasoningHard problems that need careful thinking30GPT-53 of 3 tests
- AgentsCarrying out multi-step tasks on its own20GPT-52 of 2 tests
- Following instructionsDoing exactly what it is asked20GPT-52 of 2 tests
- FactsGetting facts right instead of making them up21GPT-52 of 3 tests
- MathCompetition and research-level math10GPT-51 of 1 test
- Long documentsFinding answers in very long texts10GPT-51 of 1 test
Math and long documents 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 K2 Instruct costs 72% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where GPT-5 pulls ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+40.8points ahead
- Recent programming contest problemsLiveCodeBench v6+33.3points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+31.4points ahead
Where Kimi K2 Instruct pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+12.5points ahead
Every test, side by side
All 64 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5
- LiveCodeBench v6GPT-5 by 33.38753.7+33.3
- SWE-bench ProGPT-5 by 14.141.827.7+14.1
- SciCodeGPT-5 by 12.242.930.7+12.2
- SWE-bench VerifiedGPT-5 by 9.174.965.8+9.1
- Terminal-Bench HardGPT-5 by 9.132.623.5+9.1
- SWE-bench MultilingualGPT-5 by 855.347.3+8
ReasoningGPT-5
- Humanity's Last ExamGPT-5 by 22.128.56.4+22.1
- GPQA DiamondGPT-5 by 8.785.476.7+8.7
- CritPtGPT-5 by 5.75.70+5.7
FactsGPT-5
- AA-Omniscience · AccuracyGPT-5 by 14.940.325.4+14.9
- AA-Omniscience · Non-hallucinationKimi K2 Instruct by 12.517.830.3+12.5
- Vectara HHEM hallucination ratelower is betterGPT-5 by 2.815.117.9+2.8
Following instructionsGPT-5
- IFBenchGPT-5 by 31.473.141.7+31.4
- Multi-ChallengeGPT-5 by 15.569.654.1+15.5
Other results46 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.
- HMMT 2025GPT-5 by 54.593.338.8+54.5
- HMMT25 no toolsGPT-5 by 54.593.338.8+54.5
- BrowseComp w/ toolsGPT-5 by 47.554.97.4+47.5
- AIME 2025GPT-5 by 45.194.649.5+45.1
- AIME25 no toolsGPT-5 by 43.694.651+43.6
- BrowseComp-ZH w/ toolsGPT-5 by 40.86322.2+40.8
- FinSearchComp-T3 (Tools-allowed)GPT-5 by 38.148.510.4+38.1
- FinSearchComp-T3 w/ toolsGPT-5 by 38.148.510.4+38.1
- FinSearchComp-globalGPT-5 by 34.463.929.5+34.4
- BrowseComp-ZHGPT-5 by 34.26328.8+34.2
- LiveCodeBenchV6 no toolsGPT-5 by 30.98756.1+30.9
- OJ-Bench (cpp)GPT-5 by 30.756.225.5+30.7
- OJ-Bench (cpp) no toolsGPT-5 by 30.756.225.5+30.7
- LiveCodeBenchGPT-5 by 29.883.553.7+29.8
- Aider-PolyglotGPT-5 by 288860+28
- Frames w/ toolsGPT-5 by 27.98658.1+27.9
- Seal-0 w/ toolsGPT-5 by 26.251.425.2+26.2
- HealthBenchGPT-5 by 23.467.243.8+23.4
- HealthBench no toolsGPT-5 by 23.467.243.8+23.4
- ArtifactsBenchGPT-5 by 18.87354.2+18.8
- AA-OmniscienceGPT-5 by 17.9-8.7-26.6+17.9
- xbench-DeepSearchGPT-5 by 16.877.861+16.8
- τ²-BenchGPT-5 by 16.682.465.8+16.6
- GAIA (text only)GPT-5 by 16.276.460.2+16.2
- Terminal-BenchGPT-5 by 13.843.830+13.8
- AIR-Bench 2024GPT-5 by 13.687.774.1+13.6
- Artificial Analysis Coding IndexGPT-5 by 11.937.825.9+11.9
- GPQA (unspecified)GPT-5 by 11.585.774.2+11.5
- τ²-Bench Telecom (AA run)GPT-5 by 11.484.873.4+11.4
- AA Agentic IndexKimi K2 Instruct by 11.226.537.7+11.2
- τ²-Bench (Retail)GPT-5 by 10.581.170.6+10.5
- vectara_avg_summary_lengthGPT-5 by 103.5 rating points162.759.2+103.5 rating
- Longform Writing eval (Kimi K2 Thinking system card)GPT-5 by 8.671.462.8+8.6
- HLE (with tools)GPT-5 by 8.335.226.9+8.3
- AA IntelligenceGPT-5 by 7.72315.3+7.7
- Tau2 airlineGPT-5 by 6.162.656.5+6.1
- MMLU-ProGPT-5 by 687.181.1+6
- vectara_factual_consistencyGPT-5 by 2.884.982.1+2.8
- MMLU-ReduxGPT-5 by 2.695.392.7+2.6
- vectara_answer_rateGPT-5 by 1.399.998.6+1.3
- XSTestKimi K2 Instruct by 1.197.198.2+1.1
- Terminal-Bench w/ simulated tools (JSON)tie43.844.5tie
- anthropic_red_teamtie99.199.3tie
- HarmBenchtie97.697.4tie
- simple_safety_teststie99.8100tie
- HumanEvaltie93.493.3tie
Questions people ask
Which is better, GPT-5 or Kimi K2 Instruct?
GPT-5 wins all seven areas where both have results: coding, agents, reasoning, facts, math, long documents and following instructions. Kimi K2 Instruct wins none, but costs 72% less.
Which is better for coding?
GPT-5. It wins 6 of the 6 coding tests both models report; Kimi K2 Instruct wins none.
Which is cheaper?
GPT-5 costs $1.25 per million input tokens and $10.00 per million output tokens; Kimi K2 Instruct costs $0.60 and $2.50. That makes Kimi K2 Instruct about 72% cheaper for the same work.
How do you compare the two?
We use the 64 benchmark tests both models have published scores on. The verdict counts the 18 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 46 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.