Claude Sonnet 4 vs Llama 3.1 Instruct 405B
Wins 6 of 6 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
Wins 0 of 6 areas
—
Claude Sonnet 4 is the stronger all-rounder.
Scores updated · 33 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 software40Claude Sonnet 44 of 4 tests
- ReasoningHard problems that need careful thinking30Claude Sonnet 43 of 4 tests · 1 tie
- FactsGetting facts right instead of making them up10Claude Sonnet 41 of 2 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own10Claude Sonnet 41 of 1 test
- Long documentsFinding answers in very long texts10Claude Sonnet 41 of 1 test
- Following instructionsDoing exactly what it is asked10Claude Sonnet 41 of 1 test
Agents, 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 tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Claude Sonnet 4 pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+48.2points ahead
- Reasons across sets of long documentsAA-LCR+45points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+31.5points ahead
Where Llama 3.1 Instruct 405B pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 33 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4
- SWE-bench VerifiedClaude Sonnet 4 by 48.272.724.5+48.2
- SWE-bench ProClaude Sonnet 4 by 31.542.711.2+31.5
- Terminal-Bench HardClaude Sonnet 4 by 24.331.16.8+24.3
- SciCodeClaude Sonnet 4 by 10.14029.9+10.1
ReasoningClaude Sonnet 4
- GPQA DiamondClaude Sonnet 4 by 26.277.751.5+26.2
- SimpleBenchClaude Sonnet 4 by 22.545.523+22.5
- Humanity's Last ExamClaude Sonnet 4 by 6.710.74+6.7
- CritPttie0.30tie
FactsClaude Sonnet 4
- AA-Omniscience · Non-hallucinationClaude Sonnet 4 by 23.370.947.6+23.3
- AA-Omniscience · Accuracytie22.723.2tie
Following instructionsClaude Sonnet 4
- IFBenchClaude Sonnet 4 by 15.754.739+15.7
Other results20 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.
- τ²-Bench Telecom (AA run)Claude Sonnet 4 by 45.664.619+45.6
- LiveCodeBenchClaude Sonnet 4 by 40.868.527.7+40.8
- HarmBenchClaude Sonnet 4 by 35.39862.7+35.3
- AIR-Bench 2024Claude Sonnet 4 by 29.788.358.6+29.7
- Artificial Analysis Coding IndexClaude Sonnet 4 by 23.137.614.5+23.1
- MATH-500 (EM)Claude Sonnet 4 by 20.29473.8+20.2
- AIME 2024Claude Sonnet 4 by 20.143.423.3+20.1
- AA-OmniscienceClaude Sonnet 4 by 17.30.2-17.1+17.3
- MMMLUClaude Sonnet 4 by 12.786.573.8+12.7
- AA IntelligenceClaude Sonnet 4 by 11.618.97.3+11.6
- AA Agentic IndexClaude Sonnet 4 by 11.317.66.3+11.3
- MMLU-ProClaude Sonnet 4 by 10.483.773.3+10.4
- MMLU-ReduxClaude Sonnet 4 by 7.493.686.2+7.4
- MMLUClaude Sonnet 4 by 2.991.588.6+2.9
- anthropic_red_teamClaude Sonnet 4 by 2.799.296.5+2.7
- bbqClaude Sonnet 4 by 2.196.694.5+2.1
- simple_safety_testsClaude Sonnet 4 by 1.210098.8+1.2
- SimpleQALlama 3.1 Instruct 405B by 1.215.917.1+1.2
- IFEvalLlama 3.1 Instruct 405B by 187.688.6+1
- XSTesttie96.695.9tie
Questions people ask
Which is better, Claude Sonnet 4 or Llama 3.1 Instruct 405B?
Claude Sonnet 4 wins all six areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. Llama 3.1 Instruct 405B wins none.
Which is better for coding?
Claude Sonnet 4. It wins 4 of the 4 coding tests both models report; Llama 3.1 Instruct 405B wins none.
How do you compare the two?
We use the 33 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 20 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.