GPT-5.2 vs GPT-5.5
Wins 0 of 8 areas
—
Wins 8 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Long documents · Following instructions
GPT-5.5 is the stronger all-rounder.GPT-5.2 is cheaper.
Scores updated · 63 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.
- CodingWriting and fixing software06GPT-5.56 of 6 tests
- ReasoningHard problems that need careful thinking06GPT-5.56 of 6 tests
- MathCompetition and research-level math05GPT-5.55 of 5 tests
- Images and chartsUnderstanding pictures, charts and video04GPT-5.54 of 4 tests
- AgentsCarrying out multi-step tasks on its own14GPT-5.54 of 5 tests
- FactsGetting facts right instead of making them up13GPT-5.53 of 4 tests
- Following instructionsDoing exactly what it is asked01GPT-5.51 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts01GPT-5.51 of 1 test
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.
GPT-5.2 costs 55% 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-5.5 pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+40.8points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+32.1points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+28.7points ahead
Where GPT-5.2 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+7.8points ahead
- Real work tasks from 44 professionsGDPVal+5.6points ahead
Every test, side by side
All 63 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.5
- SWE-bench ProGPT-5.5 by 28.729.958.6+28.7
- Terminal-Bench HardGPT-5.5 by 13.64760.6+13.6
- LMArena · WebDevGPT-5.5 by 97 rating points14151512+97 rating
- LiveBench · CodingGPT-5.5 by 6.176.182.2+6.1
- LiveBench · Agentic CodingGPT-5.5 by 3.750.354+3.7
- SciCodeGPT-5.5 by 3.752.155.8+3.7
AgentsGPT-5.5
- τ-Bench V3 · BankingGPT-5.5 by 27.911.139+27.9
- BrowseCompGPT-5.5 by 18.665.884.4+18.6
- AA ApexAgentsGPT-5.5 by 14.72337.7+14.7
- MCP AtlasGPT-5.5 by 14.760.675.3+14.7
- GDPValGPT-5.2 by 5.648.342.7+5.6
ReasoningGPT-5.5
- ARC-AGI-2GPT-5.5 by 32.152.985+32.1
- SimpleBenchGPT-5.5 by 23.245.869+23.2
- CritPtGPT-5.5 by 15.511.627.1+15.5
- Humanity's Last ExamGPT-5.5 by 8.137.745.8+8.1
- LiveBench · ReasoningGPT-5.5 by 6.583.289.7+6.5
- GPQA DiamondGPT-5.5 by 3.290.393.5+3.2
FactsGPT-5.5
- SimpleQA VerifiedGPT-5.5 by 25.937.163+25.9
- AA-Omniscience · AccuracyGPT-5.5 by 13.744.358+13.7
- AA-Omniscience · Non-hallucinationGPT-5.2 by 7.818.811+7.8
- Vectara HHEM hallucination ratelower is betterGPT-5.5 by 1.510.89.3+1.5
Images and chartsGPT-5.5
- OCRBenchv2GPT-5.5 by 10.650.561.1+10.6
- MMMU-ProGPT-5.5 by 5.374.679.9+5.3
- LMArena · VisionGPT-5.5 by 46 rating points12461292+46 rating
- MathVistaGPT-5.5 by 1.482.884.2+1.4
MathGPT-5.5
- FrontierMath Tier 4GPT-5.5 by 40.831.772.5+40.8
- FrontierMath Tiers 1-3 (v2)GPT-5.5 by 17.967.485.3+17.9
- LiveBench · MathematicsGPT-5.5 by 2.793.295.9+2.7
- AIME 2026GPT-5.5 by 1.798.3100+1.7
- HMMT Feb. 2026GPT-5.5 by 1.59798.5+1.5
Following instructionsGPT-5.5
- LiveBench · Instruction FollowingGPT-5.5 by 8.961.870.7+8.9
- IFBenchtie75.475.9tie
Other results30 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.
- GDPval-AA (Elo)GPT-5.5 by 307 rating points14621769+307 rating
- Terminal-Bench 2.0GPT-5.5 by 28.75482.7+28.7
- Artificial Analysis Coding IndexGPT-5.5 by 26.248.774.9+26.2
- AA Agentic IndexGPT-5.2 by 22.960.237.3+22.9
- AA-OmniscienceGPT-5.5 by 21.4-0.920.5+21.4
- frontiermath_tier_4_v1GPT-5.5 by 16.618.835.4+16.6
- ToolathlonGPT-5.5 by 9.346.355.6+9.3
- τ²-Bench Telecom (AA run)GPT-5.5 by 9.184.893.9+9.1
- ARC-AGI-1GPT-5.5 by 8.886.295+8.8
- matharena_visual_math_overallGPT-5.5 by 8.486.594.9+8.4
- AA IntelligenceGPT-5.5 by 830.438.4+8
- livebench_languageGPT-5.5 by 7.679.887.4+7.6
- HLE (with tools)GPT-5.5 by 6.745.552.2+6.7
- WorldVQAGPT-5.5 by 6.62834.6+6.6
- HealthBench Professional length-adjustedGPT-5.5 by 5.945.951.8+5.9
- LiveBenchGPT-5.5 by 5.974.880.7+5.9
- vectara_avg_summary_lengthGPT-5.2 by 56.7 rating points186.3129.6+56.7 rating
- FrontierMath (overall)GPT-5.2 by 4.940.335.4+4.9
- livebench_data_analysisGPT-5.5 by 3.478.281.6+3.4
- HealthBench Hard length-adjustedGPT-5.2 by 2.834.331.5+2.8
- SimpleVQAGPT-5.5 by 2.855.858.6+2.8
- vectara_factual_consistencyGPT-5.5 by 1.589.290.7+1.5
- HealthBench Consensus length-adjustedGPT-5.5 by 1.294.495.6+1.2
- HMMT Nov. 2025tie95.896.5tie
- HealthBench length-adjustedtie56.856.5tie
- Image input eval - extremism (not_unsafe)tie11tie
- Image input eval - harms-erotic (not_unsafe)tie11tie
- Image input eval - hate (not_unsafe)tie11tie
- Image input eval - self-harm (not_unsafe)tie11tie
- vectara_answer_ratetie100100tie
Questions people ask
Which is better, GPT-5.2 or GPT-5.5?
GPT-5.5 wins all eight areas we test: coding, agents, reasoning, facts, images and charts, math, long documents and following instructions. GPT-5.2 wins none, but costs 55% less.
Which is better for coding?
GPT-5.5. It wins 6 of the 6 coding tests both models report; GPT-5.2 wins none.
Which is cheaper?
GPT-5.2 costs $1.75 per million input tokens and $14.00 per million output tokens; GPT-5.5 costs $5.00 and $30.00. That makes GPT-5.2 about 55% cheaper for the same work.
How do you compare the two?
We use the 63 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 30 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.