Claude 3.5 Haiku vs Llama 3.1 Instruct 405B
Wins 2 of 6 areas
Long documents · Following instructions
Wins 2 of 6 areas
Coding · Reasoning
The two are evenly matched.Claude 3.5 Haiku is better at long documents and following instructions; Llama 3.1 Instruct 405B at coding and reasoning.
Scores updated · 27 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.
- Long documentsFinding answers in very long texts10Claude 3.5 Haiku1 of 1 test
- Following instructionsDoing exactly what it is asked10Claude 3.5 Haiku1 of 1 test
- CodingWriting and fixing software12Llama 3.1 Instruct 405B2 of 3 tests
- ReasoningHard problems that need careful thinking01Llama 3.1 Instruct 405B1 of 3 tests · 2 ties
- AgentsCarrying out multi-step tasks on its own00Even0 each · 1 tie
- FactsGetting facts right instead of making them up11Even1 each
Agents, long documents and following instructions rest on a single test each.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Claude 3.5 Haiku pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+16.1points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+11.3points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+3.8points ahead
Where Llama 3.1 Instruct 405B pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+10.7points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+10points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+4.5points ahead
Every test, side by side
All 27 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingLlama 3.1 Instruct 405B
- SWE-bench VerifiedClaude 3.5 Haiku by 16.140.624.5+16.1
- Terminal-Bench HardLlama 3.1 Instruct 405B by 4.52.36.8+4.5
- SciCodeLlama 3.1 Instruct 405B by 2.527.429.9+2.5
ReasoningLlama 3.1 Instruct 405B
- GPQA DiamondLlama 3.1 Instruct 405B by 10.740.851.5+10.7
- Humanity's Last Examtie3.64tie
- CritPttie00tie
FactsEven
- AA-Omniscience · Non-hallucinationClaude 3.5 Haiku by 11.358.947.6+11.3
- AA-Omniscience · AccuracyLlama 3.1 Instruct 405B by 1013.223.2+10
Following instructionsClaude 3.5 Haiku
- IFBenchClaude 3.5 Haiku by 3.842.839+3.8
Other results16 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.
- MMLULlama 3.1 Instruct 405B by 21.567.188.6+21.5
- MBPPClaude 3.5 Haiku by 17.285.668.4+17.2
- GSM8KLlama 3.1 Instruct 405B by 15.381.596.8+15.3
- MATHClaude 3.5 Haiku by 13.487.273.8+13.4
- naturalquestions_closedbookLlama 3.1 Instruct 405B by 11.234.445.6+11.2
- OpenBookQALlama 3.1 Instruct 405B by 8.685.494+8.6
- MMLU-ProLlama 3.1 Instruct 405B by 8.36573.3+8.3
- MGSMLlama 3.1 Instruct 405B by 685.691.6+6
- τ²-Bench Telecom (AA run)Claude 3.5 Haiku by 5.624.619+5.6
- AA-OmniscienceLlama 3.1 Instruct 405B by 5.4-22.5-17.1+5.4
- AA Agentic IndexLlama 3.1 Instruct 405B by 4.41.96.3+4.4
- DROPLlama 3.1 Instruct 405B by 1.783.184.8+1.7
- AA IntelligenceClaude 3.5 Haiku by 1.68.97.3+1.6
- Artificial Analysis Coding IndexClaude 3.5 Haiku by 1.415.914.5+1.4
- NarrativeQAClaude 3.5 Haiku by 1.476.374.9+1.4
- HumanEvaltie88.189tie
Questions people ask
Which is better, Claude 3.5 Haiku or Llama 3.1 Instruct 405B?
Claude 3.5 Haiku and Llama 3.1 Instruct 405B each win two of the six areas where both have results. Claude 3.5 Haiku wins long documents and following instructions; Llama 3.1 Instruct 405B wins coding and reasoning. They are level on agents and facts.
Which is better for coding?
Llama 3.1 Instruct 405B. 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 27 benchmark tests both models have published scores on. The verdict counts the 11 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 16 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.