GPT-5.3 Codex vs GPT-5.6 Sol
Wins 1 of 7 areas
Following instructions
Wins 4 of 7 areas
Coding · Agents · Reasoning · Images and charts
GPT-5.6 Sol wins more areas, narrowly.GPT-5.3 Codex is cheaper and better at following instructions.
Scores updated · 20 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 software04GPT-5.6 Sol4 of 4 tests
- ReasoningHard problems that need careful thinking03GPT-5.6 Sol3 of 3 tests
- AgentsCarrying out multi-step tasks on its own02GPT-5.6 Sol2 of 2 tests
- Images and chartsUnderstanding pictures, charts and video01GPT-5.6 Sol1 of 1 test
- Following instructionsDoing exactly what it is asked10GPT-5.3 Codex1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
- Long documentsFinding answers in very long texts00Even0 each · 1 tie
Images and charts, long documents and following instructions 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.
GPT-5.3 Codex costs 34% 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
- Completes tasks by operating a computer desktopOSWorld-Verified+18.3points ahead
- Research-level physics problemsCritPt+15.4points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+12.9points ahead
Where GPT-5.3 Codex pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+3points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+2.7points ahead
Every test, side by side
All 20 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 209 rating points14091618+209 rating
- Terminal-Bench HardGPT-5.6 Sol by 12.95365.9+12.9
- SWE-bench ProGPT-5.6 Sol by 7.856.864.6+7.8
- SciCodeGPT-5.6 Sol by 3.953.257.1+3.9
AgentsGPT-5.6 Sol
- OSWorld-VerifiedGPT-5.6 Sol by 18.364.783+18.3
- GDPValGPT-5.6 Sol by 6.64955.6+6.6
ReasoningGPT-5.6 Sol
- CritPtGPT-5.6 Sol by 15.416.932.3+15.4
- Humanity's Last ExamGPT-5.6 Sol by 742.549.5+7
- GPQA DiamondGPT-5.6 Sol by 2.691.594.1+2.6
FactsEven
- AA-Omniscience · AccuracyGPT-5.6 Sol by 6.552.959.4+6.5
- AA-Omniscience · Non-hallucinationGPT-5.3 Codex by 310.87.8+3
Following instructionsGPT-5.3 Codex
- IFBenchGPT-5.3 Codex by 2.775.472.7+2.7
Other results6 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.
- HiL-Bench (Tools-allowed)GPT-5.6 Sol by 284.332.3+28
- Artificial Analysis Coding IndexGPT-5.6 Sol by 24.353.177.4+24.3
- AA IntelligenceGPT-5.6 Sol by 14.532.547+14.5
- AA-OmniscienceGPT-5.6 Sol by 11.110.922+11.1
- AA Agentic IndexGPT-5.3 Codex by 1060.550.5+10
- τ²-Bench Telecom (AA run)tie8685.1tie
Questions people ask
Which is better, GPT-5.3 Codex or GPT-5.6 Sol?
GPT-5.6 Sol wins four of the seven areas where both have results: coding, agents, reasoning and images and charts. GPT-5.3 Codex wins following instructions, and costs 34% less. They are level on facts and long documents.
Which is better for coding?
GPT-5.6 Sol. It wins 4 of the 4 coding tests both models report; GPT-5.3 Codex wins none.
Which is cheaper?
GPT-5.3 Codex costs $1.75 per million input tokens and $14.00 per million output tokens; GPT-5.6 Sol costs $4.00 and $20.00. That makes GPT-5.3 Codex about 34% cheaper for the same work.
How do you compare the two?
We use the 20 benchmark tests both models have published scores on. The verdict counts the 14 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 6 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.