Claude 3.5 Haiku vs GPT-4o
Wins 1 of 7 areas
Following instructions
Wins 4 of 7 areas
Coding · Facts · Images and charts · Long documents
GPT-4o wins more areas, narrowly.Claude 3.5 Haiku is better at following instructions.
Scores updated · 31 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.
- FactsGetting facts right instead of making them up03GPT-4o3 of 3 tests
- CodingWriting and fixing software12GPT-4o2 of 3 tests
- Images and chartsUnderstanding pictures, charts and video01GPT-4o1 of 1 test
- Long documentsFinding answers in very long texts01GPT-4o1 of 1 test
- Following instructionsDoing exactly what it is asked10Claude 3.5 Haiku1 of 1 test
- AgentsCarrying out multi-step tasks on its own00Even0 each · 1 tie
- ReasoningHard problems that need careful thinking11Even1 each · 1 tie
Agents, 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.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where GPT-4o pulls ahead
- Reasons across sets of long documentsAA-LCR+28.7points ahead
- Short factual questions, answered correctlySimpleQA Verified+19.3points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+11.8points ahead
Where Claude 3.5 Haiku pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+7.4points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+6.8points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+1.8points ahead
Every test, side by side
All 31 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-4o
- SWE-bench VerifiedClaude 3.5 Haiku by 7.440.633.2+7.4
- Terminal-Bench HardGPT-4o by 62.38.3+6
- SciCodeGPT-4o by 5.927.433.3+5.9
ReasoningEven
- GPQA DiamondGPT-4o by 11.840.852.6+11.8
- Humanity's Last ExamClaude 3.5 Haiku by 1.83.61.8+1.8
- CritPttie00tie
FactsGPT-4o
- SimpleQA VerifiedGPT-4o by 19.36.726+19.3
- AA-Omniscience · AccuracyGPT-4o by 10.513.223.7+10.5
- AA-Omniscience · Non-hallucinationGPT-4o by 3.258.962.1+3.2
Following instructionsClaude 3.5 Haiku
- IFBenchClaude 3.5 Haiku by 6.842.836+6.8
Other results18 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.
- TAU-bench (airline)GPT-4o by 2022.842.8+20
- naturalquestions_closedbookGPT-4o by 15.234.449.6+15.2
- AA-OmniscienceGPT-4o by 12-22.5-10.5+12
- OpenBookQAGPT-4o by 11.485.496.8+11.4
- MMLU-ProGPT-4o by 9.76574.7+9.7
- GSM8KGPT-4o by 9.481.590.9+9.4
- TAU-bench (retail)GPT-4o by 9.35160.3+9.3
- Artificial Analysis Coding IndexGPT-4o by 8.315.924.2+8.3
- MMLUGPT-4o by 6.767.173.8+6.7
- AA Agentic IndexGPT-4o by 6.51.98.4+6.5
- MGSMGPT-4o by 4.985.690.5+4.9
- τ²-Bench Telecom (AA run)GPT-4o by 4.324.628.9+4.3
- NarrativeQAGPT-4o by 3.276.379.5+3.2
- Aider-PolyglotGPT-4o by 2.72830.7+2.7
- HumanEvalGPT-4o by 2.188.190.2+2.1
- MATHClaude 3.5 Haiku by 1.987.285.3+1.9
- AA IntelligenceClaude 3.5 Haiku by 1.68.97.3+1.6
- DROPtie83.183.4tie
Questions people ask
Which is better, Claude 3.5 Haiku or GPT-4o?
GPT-4o wins four of the seven areas where both have results: coding, facts, images and charts and long documents. Claude 3.5 Haiku wins following instructions. They are level on agents and reasoning.
Which is better for coding?
GPT-4o. It wins 2 of the 3 coding tests both models report; Claude 3.5 Haiku wins 1.
How do you compare the two?
We use the 31 benchmark tests both models have published scores on. The verdict counts the 13 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 18 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.