GPT-5.4 vs GPT-5.6 Sol
Wins 0 of 8 areas
—
Wins 7 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Long documents
GPT-5.6 Sol is the stronger all-rounder.GPT-5.4 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.
- AgentsCarrying out multi-step tasks on its own07GPT-5.6 Sol7 of 7 tests
- CodingWriting and fixing software06GPT-5.6 Sol6 of 7 tests · 1 tie
- ReasoningHard problems that need careful thinking06GPT-5.6 Sol6 of 6 tests
- MathCompetition and research-level math03GPT-5.6 Sol3 of 4 tests · 1 tie
- FactsGetting facts right instead of making them up12GPT-5.6 Sol2 of 4 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video12GPT-5.6 Sol2 of 3 tests
- Long documentsFinding answers in very long texts01GPT-5.6 Sol1 of 1 test
- 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.
GPT-5.4 costs 27% 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-5.6 Sol pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+34points ahead
- Short factual questions, answered correctlySimpleQA Verified+24.6points ahead
- Finds the root cause of IT system incidentsAA IT-Bench SRE+21.7points ahead
Where GPT-5.4 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+1.2points 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-5.6 Sol
- LMArena · WebDevGPT-5.6 Sol by 153 rating points14651618+153 rating
- Terminal-Bench 2.1GPT-5.6 Sol by 9.778.388+9.7
- Terminal-Bench HardGPT-5.6 Sol by 8.357.665.9+8.3
- SWE-bench ProGPT-5.6 Sol by 6.957.764.6+6.9
- LiveBench · CodingGPT-5.6 Sol by 6.477.583.9+6.4
- LiveBench · Agentic CodingGPT-5.6 Sol by 2.453.856.2+2.4
- SciCodetie56.657.1tie
AgentsGPT-5.6 Sol
- AA IT-Bench SREGPT-5.6 Sol by 21.734.556.2+21.7
- GDPValGPT-5.6 Sol by 18.237.455.6+18.2
- MCP AtlasGPT-5.6 Sol by 16.467.283.6+16.4
- OSWorld-VerifiedGPT-5.6 Sol by 87583+8
- BrowseCompGPT-5.6 Sol by 7.782.790.4+7.7
- AA ApexAgentsGPT-5.6 Sol by 6.633.339.9+6.6
- τ-Bench V3 · BankingGPT-5.6 Sol by 4.739.644.3+4.7
ReasoningGPT-5.6 Sol
- ARC-AGI-2GPT-5.6 Sol by 19.273.392.5+19.2
- CritPtGPT-5.6 Sol by 8.923.432.3+8.9
- ARC-AGI-3GPT-5.6 Sol by 7.60.27.8+7.6
- Humanity's Last ExamGPT-5.6 Sol by 5.843.749.5+5.8
- LiveBench · ReasoningGPT-5.6 Sol by 3.688.191.7+3.6
- GPQA DiamondGPT-5.6 Sol by 2.19294.1+2.1
FactsGPT-5.6 Sol
- SimpleQA VerifiedGPT-5.6 Sol by 24.645.169.7+24.6
- AA-Omniscience · AccuracyGPT-5.6 Sol by 8.550.959.4+8.5
- Vectara HHEM hallucination ratelower is betterGPT-5.4 by 5.4712.4+5.4
- AA-Omniscience · Non-hallucinationtie8.37.8tie
Images and chartsGPT-5.6 Sol
- MMMU-ProGPT-5.6 Sol by 578.483.4+5
- CharXiv (RQ)GPT-5.6 Sol by 382.885.8+3
- LMArena · VisionGPT-5.4 by 23 rating points13021279+23 rating
MathGPT-5.6 Sol
- FrontierMath Tier 4GPT-5.6 Sol by 344983+34
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Sol by 10.578.689.1+10.5
- LiveBench · MathematicsGPT-5.6 Sol by 294.296.2+2
- AIME 2026tie99.299.9tie
Following instructionsEven
- LiveBench · Instruction FollowingGPT-5.6 Sol by 1.770.271.8+1.7
- IFBenchGPT-5.4 by 1.273.972.7+1.2
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.
- FrontierMath (overall)GPT-5.6 Sol by 41.447.689+41.4
- HiL-Bench (Tools-allowed)GPT-5.6 Sol by 22.69.732.3+22.6
- DeepSWE 1.1GPT-5.6 Sol by 215273+21
- CyberGymGPT-5.6 Sol by 18.266.384.5+18.2
- AA-OmniscienceGPT-5.6 Sol by 16.25.822+16.2
- HLE (with tools)GPT-5.6 Sol by 12.55264.5+12.5
- HealthBench Professional length-adjustedGPT-5.6 Sol by 12.448.160.5+12.4
- OfficeQA ProGPT-5.6 Sol by 12.151.163.2+12.1
- AA IntelligenceGPT-5.6 Sol by 83947+8
- AA Agentic IndexGPT-5.6 Sol by 6.344.250.5+6.3
- Artificial Analysis Coding IndexGPT-5.6 Sol by 6.371.177.4+6.3
- vectara_factual_consistencyGPT-5.4 by 5.49387.6+5.4
- livebench_languageGPT-5.6 Sol by 5.182.687.7+5.1
- vectara_avg_summary_lengthGPT-5.6 Sol by 49.4 rating points81.7131.1+49.4 rating
- Hard-negative protein binding predictionGPT-5.6 Sol by 4.13.57.6+4.1
- HealthBench Hard length-adjustedGPT-5.6 Sol by 429.133.1+4
- ToolathlonGPT-5.6 Sol by 3.454.658+3.4
- HealthBench length-adjustedGPT-5.6 Sol by 35457+3
- ARC-AGI-1GPT-5.6 Sol by 2.893.796.5+2.8
- charxiv_rq_with_pythonGPT-5.4 by 2.29087.8+2.2
- τ²-Bench Telecom (AA run)GPT-5.4 by 287.185.1+2
- DNA sequence design for transcription factor bindingtie12.813.7tie
- HealthBench Consensus length-adjustedtie96.395.5tie
- vectara_answer_ratetie99.999.1tie
- livebench_data_analysistie79.379.8tie
- 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
Questions people ask
Which is better, GPT-5.4 or GPT-5.6 Sol?
GPT-5.6 Sol wins seven of the eight areas we test: coding, agents, reasoning, facts, images and charts, math and long documents. GPT-5.4 wins none, but costs 27% less. They are level on following instructions.
Which is better for coding?
GPT-5.6 Sol. It wins 6 of the 7 coding tests both models report; GPT-5.4 wins none, 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; GPT-5.6 Sol costs $4.00 and $20.00. That makes GPT-5.4 about 27% 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 34 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.