GPT-4o vs Llama 3.1 Instruct 405B
Wins 3 of 6 areas
Coding · Facts · Long documents
Wins 1 of 6 areas
Following instructions
GPT-4o wins more areas, narrowly.Llama 3.1 Instruct 405B is better at following instructions.
Scores updated · 54 tests both models report · How we compare
Where each one wins
Tests won in each of the six 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 software30GPT-4o3 of 3 tests
- Long documentsFinding answers in very long texts20GPT-4o2 of 2 tests
- FactsGetting facts right instead of making them up10GPT-4o1 of 2 tests · 1 tie
- Following instructionsDoing exactly what it is asked01Llama 3.1 Instruct 405B1 of 1 test
- AgentsCarrying out multi-step tasks on its own00Even0 each · 1 tie
- ReasoningHard problems that need careful thinking11Even1 each · 1 tie
Agents 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 tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where GPT-4o pulls ahead
- Reasons across sets of long documentsAA-LCR+30.7points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+14.5points ahead
- Questions about very long textsLongBench v2+12points ahead
Where Llama 3.1 Instruct 405B pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+3points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+2.2points ahead
Every test, side by side
All 54 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-4o
- SWE-bench VerifiedGPT-4o by 8.733.224.5+8.7
- SciCodeGPT-4o by 3.433.329.9+3.4
- Terminal-Bench HardGPT-4o by 1.58.36.8+1.5
ReasoningEven
- Humanity's Last ExamLlama 3.1 Instruct 405B by 2.21.84+2.2
- GPQA DiamondGPT-4o by 1.152.651.5+1.1
- CritPttie00tie
FactsGPT-4o
- AA-Omniscience · Non-hallucinationGPT-4o by 14.562.147.6+14.5
- AA-Omniscience · Accuracytie23.723.2tie
Long documentsGPT-4o
- AA-LCRGPT-4o by 30.75625.3+30.7
- LongBench v2GPT-4o by 1248.136.1+12
Following instructionsLlama 3.1 Instruct 405B
- IFBenchLlama 3.1 Instruct 405B by 33639+3
Other results42 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.
- CodeforcesGPT-4o by 733.7 rating points75925.3+733.7 rating
- SimpleQAGPT-4o by 21.138.217.1+21.1
- HarmBenchGPT-4o by 20.282.962.7+20.2
- AlpacaEval 2 LCGPT-4o by 18.257.539.3+18.2
- MMLULlama 3.1 Instruct 405B by 14.873.888.6+14.8
- AIME 2024Llama 3.1 Instruct 405B by 149.323.3+14
- MATHGPT-4o by 11.585.373.8+11.5
- LiveCodeBenchGPT-4o by 10.638.327.7+10.6
- FRAMES (Acc.)GPT-4o by 10.580.570+10.5
- Arena HardGPT-4o by 1079.369.3+10
- τ²-Bench Telecom (AA run)GPT-4o by 9.928.919+9.9
- Artificial Analysis Coding IndexGPT-4o by 9.724.214.5+9.7
- Aider-Edit (Acc.)GPT-4o by 972.963.9+9
- Chinese SimpleQA (C-SimpleQA)GPT-4o by 8.358.750.4+8.3
- IFEvalLlama 3.1 Instruct 405B by 7.68188.6+7.6
- MMMLUGPT-4o by 7.681.473.8+7.6
- AA-OmniscienceGPT-4o by 6.6-10.5-17.1+6.6
- GSM8KLlama 3.1 Instruct 405B by 5.990.996.8+5.9
- DROP (3-shot F1)Llama 3.1 Instruct 405B by 583.788.7+5
- CLUEWSCGPT-4o by 4.987.983+4.9
- NarrativeQAGPT-4o by 4.679.574.9+4.6
- naturalquestions_closedbookGPT-4o by 449.645.6+4
- AIR-Bench 2024GPT-4o by 3.862.458.6+3.8
- C-EvalGPT-4o by 3.57672.5+3.5
- HumanEval-Mul (Pass@1)GPT-4o by 3.380.577.2+3.3
- DROP (F1)GPT-4o by 3.289.286+3.2
- MBPP+ (EvalPlus-augmented)GPT-4o by 3.276.273+3.2
- OpenBookQAGPT-4o by 2.896.894+2.8
- anthropic_red_teamGPT-4o by 2.699.196.5+2.6
- IFEval (avg)Llama 3.1 Instruct 405B by 2.384.186.4+2.3
- AA Agentic IndexGPT-4o by 2.18.46.3+2.1
- MMLU-ReduxGPT-4o by 1.88886.2+1.8
- DROPLlama 3.1 Instruct 405B by 1.483.484.8+1.4
- MMLU-ProGPT-4o by 1.474.773.3+1.4
- XSTestGPT-4o by 1.497.395.9+1.4
- HumanEvalGPT-4o by 1.290.289+1.2
- MGSMLlama 3.1 Instruct 405B by 1.190.591.6+1.1
- MATH-500 (EM)tie74.673.8tie
- bbqtie95.194.5tie
- simple_safety_teststie98.598.8tie
- MT-Benchtie8.78.5tie
- AA Intelligencetie7.37.3tie
Questions people ask
Which is better, GPT-4o or Llama 3.1 Instruct 405B?
GPT-4o wins three of the six areas where both have results: coding, facts and long documents. Llama 3.1 Instruct 405B wins following instructions. They are level on agents and reasoning.
Which is better for coding?
GPT-4o. It wins 3 of the 3 coding tests both models report; Llama 3.1 Instruct 405B wins none.
How do you compare the two?
We use the 54 benchmark tests both models have published scores on. The verdict counts the 12 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 42 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.