GPT-5.6 Sol wins more areas, narrowly.
Both rank among the ten best models we track in reasoning, coding, agents and math.
Scores updated · 79 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 own08GPT-5.6 Sol8 of 8 tests
- CodingWriting and fixing software07GPT-5.6 Sol7 of 7 tests
- ReasoningHard problems that need careful thinking15GPT-5.6 Sol5 of 7 tests · 1 tie
- MathCompetition and research-level math02GPT-5.6 Sol2 of 4 tests · 2 ties
- FactsGetting facts right instead of making them up22Even2 each
- Images and chartsUnderstanding pictures, charts and video11Even1 each
- Long documentsFinding answers in very long texts00Even0 each · 1 tie
- 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-5.6 Sol costs 31% 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-5.6 Sol pulls ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+25.3points ahead
- Real work tasks from 44 professionsGDPVal+12.9points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+10.5points ahead
Where GPT-5.5 pulls ahead
- Common-sense trick questionsSimpleBench+4.2points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+3.2points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+3.2points ahead
Every test, side by side
All 79 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 108 rating points15121620+108 rating
- SWE-bench ProGPT-5.6 Sol by 658.664.6+6
- Terminal-Bench HardGPT-5.6 Sol by 5.360.665.9+5.3
- Terminal-Bench 2.1GPT-5.6 Sol by 3.784.388+3.7
- LiveBench · Agentic CodingGPT-5.6 Sol by 2.25456.2+2.2
- LiveBench · CodingGPT-5.6 Sol by 1.782.283.9+1.7
- SciCodeGPT-5.6 Sol by 1.355.857.1+1.3
AgentsGPT-5.6 Sol
- Terminal-Bench 4.0GPT-5.6 Sol by 25.314.639.9+25.3
- GDPValGPT-5.6 Sol by 12.942.755.6+12.9
- AA IT-Bench SREGPT-5.6 Sol by 10.445.856.2+10.4
- MCP AtlasGPT-5.6 Sol by 8.375.383.6+8.3
- BrowseCompGPT-5.6 Sol by 684.490.4+6
- τ-Bench V3 · BankingGPT-5.6 Sol by 5.33944.3+5.3
- OSWorld-VerifiedGPT-5.6 Sol by 4.378.783+4.3
- AA ApexAgentsGPT-5.6 Sol by 2.237.739.9+2.2
ReasoningGPT-5.6 Sol
- ARC-AGI-2GPT-5.6 Sol by 7.58592.5+7.5
- ARC-AGI-3GPT-5.6 Sol by 7.40.47.8+7.4
- CritPtGPT-5.6 Sol by 5.227.132.3+5.2
- SimpleBenchGPT-5.5 by 4.26964.8+4.2
- Humanity's Last ExamGPT-5.6 Sol by 3.745.849.5+3.7
- LiveBench · ReasoningGPT-5.6 Sol by 289.791.7+2
- GPQA Diamondtie93.594.1tie
FactsEven
- SimpleQA VerifiedGPT-5.6 Sol by 6.76369.7+6.7
- AA-Omniscience · Non-hallucinationGPT-5.5 by 3.2117.8+3.2
- Vectara HHEM hallucination ratelower is betterGPT-5.5 by 3.19.312.4+3.1
- AA-Omniscience · AccuracyGPT-5.6 Sol by 1.45859.4+1.4
Images and chartsEven
- MMMU-ProGPT-5.6 Sol by 3.579.983.4+3.5
- LMArena · VisionGPT-5.5 by 13 rating points12921279+13 rating
MathGPT-5.6 Sol
- FrontierMath Tier 4GPT-5.6 Sol by 10.572.583+10.5
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Sol by 3.885.389.1+3.8
- LiveBench · Mathematicstie95.996.2tie
- AIME 2026tie10099.9tie
Following instructionsEven
- IFBenchGPT-5.5 by 3.275.972.7+3.2
- LiveBench · Instruction FollowingGPT-5.6 Sol by 1.270.771.8+1.2
Other results44 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 53.635.489+53.6
- Program BenchGPT-5.5 by 45.870.825+45.8
- SWE-MarathonGPT-5.6 Sol by 30.51242.5+30.5
- Agents' Last ExamGPT-5.6 Sol by 26.126.652.7+26.1
- GDPval-AA v2 EloGPT-5.6 Sol by 219 rating points14911710+219 rating
- GDPval-AA v2GPT-5.6 Sol by 202 rating points15091711+202 rating
- AA Agentic IndexGPT-5.6 Sol by 13.237.350.5+13.2
- HLE (with tools)GPT-5.6 Sol by 12.352.264.5+12.3
- MLS Bench Litelower is betterGPT-5.5 by 10.735.546.2+10.7
- ResearchRubricsGPT-5.6 Sol by 9.86473.8+9.8
- OfficeQA ProGPT-5.6 Sol by 9.154.163.2+9.1
- τ²-Bench Telecom (AA run)GPT-5.5 by 8.893.985.1+8.8
- HealthBench ProfessionalGPT-5.6 Sol by 8.751.860.5+8.7
- HealthBench Professional length-adjustedGPT-5.6 Sol by 8.751.860.5+8.7
- AA IntelligenceGPT-5.6 Sol by 8.638.447+8.6
- PostTrainBenchGPT-5.6 Sol by 7.828.436.2+7.8
- HiL-Bench (Tools-allowed)GPT-5.5 by 7.439.732.3+7.4
- JobBenchGPT-5.6 Sol by 7.138.345.4+7.1
- Hard-negative protein binding predictionGPT-5.6 Sol by 6.11.57.6+6.1
- DeepSWE 1.1GPT-5.6 Sol by 66773+6
- GDP (Surge AI)GPT-5.6 Sol by 5.824.930.7+5.8
- SpreadsheetBench 2GPT-5.6 Sol by 3.329.132.4+3.3
- vectara_factual_consistencyGPT-5.5 by 3.190.787.6+3.1
- CyberGymGPT-5.6 Sol by 2.781.884.5+2.7
- DeepSWEGPT-5.6 Sol by 2.77072.7+2.7
- Artificial Analysis Coding IndexGPT-5.6 Sol by 2.574.977.4+2.5
- ToolathlonGPT-5.6 Sol by 2.455.658+2.4
- livebench_data_analysisGPT-5.5 by 1.881.679.8+1.8
- HealthBench Hard length-adjustedGPT-5.6 Sol by 1.631.533.1+1.6
- AA-OmniscienceGPT-5.6 Sol by 1.520.522+1.5
- ARC-AGI-1GPT-5.6 Sol by 1.59596.5+1.5
- Toolathlon VerifiedGPT-5.6 Sol by 1.473.574.9+1.4
- vectara_answer_ratetie10099.1tie
- HealthBenchtie56.557tie
- HealthBench length-adjustedtie56.557tie
- livebench_languagetie87.487.7tie
- vectara_avg_summary_lengthtie129.6131.1tie
- DNA sequence design for transcription factor bindingtie13.813.7tie
- HealthBench Consensus length-adjustedtie95.695.5tie
- 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
- MCP Mark Verifiedtie92.992.9tie
Questions people ask
Which is better, GPT-5.5 or GPT-5.6 Sol?
GPT-5.6 Sol wins four of the eight areas we test: coding, agents, reasoning and math. GPT-5.5 wins none. They are level on facts, images and charts, long documents and following instructions.
Which is better for coding?
GPT-5.6 Sol. It wins 7 of the 7 coding tests both models report; GPT-5.5 wins none.
Which is cheaper?
GPT-5.5 costs $5.00 per million input tokens and $30.00 per million output tokens; GPT-5.6 Sol costs $4.00 and $20.00. That makes GPT-5.6 Sol about 31% cheaper for the same work.
How do you compare the two?
We use the 79 benchmark tests both models have published scores on. The verdict counts the 35 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 44 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.