Claude Opus 4.6 vs GPT-5.3 Codex
Wins 1 of 7 areas
Agents
Wins 5 of 7 areas
Coding · Reasoning · Images and charts · Long documents · Following instructions
GPT-5.3 Codex is the stronger all-rounder.Claude Opus 4.6 is better at agents.
Scores updated · 23 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.
- ReasoningHard problems that need careful thinking03GPT-5.3 Codex3 of 3 tests
- CodingWriting and fixing software13GPT-5.3 Codex3 of 4 tests
- Images and chartsUnderstanding pictures, charts and video01GPT-5.3 Codex1 of 1 test
- Long documentsFinding answers in very long texts01GPT-5.3 Codex1 of 1 test
- Following instructionsDoing exactly what it is asked01GPT-5.3 Codex1 of 1 test
- AgentsCarrying out multi-step tasks on its own20Claude Opus 4.62 of 2 tests
- FactsGetting facts right instead of making them up11Even1 each
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 48% 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-5.3 Codex pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+22.3points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+6.8points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+5.9points ahead
Where Claude Opus 4.6 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+26.4points ahead
- Completes tasks by operating a computer desktopOSWorld-Verified+8points ahead
- Real work tasks from 44 professionsGDPVal+6.9points ahead
Every test, side by side
All 23 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.3 Codex
- LMArena · WebDevClaude Opus 4.6 by 137 rating points15461409+137 rating
- Terminal-Bench HardGPT-5.3 Codex by 6.846.253+6.8
- SWE-bench ProGPT-5.3 Codex by 3.453.456.8+3.4
- SciCodeGPT-5.3 Codex by 1.351.953.2+1.3
AgentsClaude Opus 4.6
- OSWorld-VerifiedClaude Opus 4.6 by 872.764.7+8
- GDPValClaude Opus 4.6 by 6.955.949+6.9
ReasoningGPT-5.3 Codex
- CritPtGPT-5.3 Codex by 4.312.616.9+4.3
- Humanity's Last ExamGPT-5.3 Codex by 2.639.942.5+2.6
- GPQA DiamondGPT-5.3 Codex by 1.989.691.5+1.9
FactsEven
- AA-Omniscience · Non-hallucinationClaude Opus 4.6 by 26.437.210.8+26.4
- AA-Omniscience · AccuracyGPT-5.3 Codex by 5.94752.9+5.9
Images and chartsGPT-5.3 Codex
- MMMU-ProGPT-5.3 Codex by 3.175.478.5+3.1
Following instructionsGPT-5.3 Codex
- IFBenchGPT-5.3 Codex by 22.353.175.4+22.3
Other results9 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)Claude Opus 4.6 by 3438.34.3+34
- Terminal-Bench 2.0GPT-5.3 Codex by 11.965.477.3+11.9
- AA Agentic IndexClaude Opus 4.6 by 7.167.660.5+7.1
- τ²-Bench Telecom (AA run)Claude Opus 4.6 by 6.192.186+6.1
- Artificial Analysis Coding IndexGPT-5.3 Codex by 548.153.1+5
- LiveBenchClaude Opus 4.6 by 3.576.372.8+3.5
- AA-OmniscienceClaude Opus 4.6 by 2.813.710.9+2.8
- metr_swaaGPT-5.3 Codex by 198.599.5+1
- AA Intelligencetie31.932.5tie
Questions people ask
Which is better, Claude Opus 4.6 or GPT-5.3 Codex?
GPT-5.3 Codex wins five of the seven areas where both have results: coding, reasoning, images and charts, long documents and following instructions. Claude Opus 4.6 wins agents. They are level on facts.
Which is better for coding?
GPT-5.3 Codex. It wins 3 of the 4 coding tests both models report; Claude Opus 4.6 wins 1.
Which is cheaper?
Claude Opus 4.6 costs $5.00 per million input tokens and $25.00 per million output tokens; GPT-5.3 Codex costs $1.75 and $14.00. That makes GPT-5.3 Codex about 48% cheaper for the same work.
How do you compare the two?
We use the 23 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 9 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.