Claude 3.5 Haiku vs Llama 3.3 70B Instruct
Wins 3 of 6 areas
Coding · Agents · Long documents
Wins 2 of 6 areas
Reasoning · Following instructions
Claude 3.5 Haiku wins more areas, narrowly.Llama 3.3 70B Instruct is better at reasoning.
Scores updated · 26 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 software20Claude 3.5 Haiku2 of 3 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own10Claude 3.5 Haiku1 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts10Claude 3.5 Haiku1 of 1 test
- ReasoningHard problems that need careful thinking01Llama 3.3 70B Instruct1 of 3 tests · 2 ties
- Following instructionsDoing exactly what it is asked01Llama 3.3 70B Instruct1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
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 Claude 3.5 Haiku pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+49.1points ahead
- Reasons across sets of long documentsAA-LCR+11.6points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+5.2points ahead
Where Llama 3.3 70B Instruct pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+9points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+5.8points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+4.3points ahead
Every test, side by side
All 26 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude 3.5 Haiku
- Terminal-Bench 2.1Claude 3.5 Haiku by 5.210.14.9+5.2
- SciCodeClaude 3.5 Haiku by 1.427.426+1.4
- Terminal-Bench Hardtie2.33tie
AgentsClaude 3.5 Haiku
- τ-Bench V3 · BankingClaude 3.5 Haiku by 4.85.81+4.8
- GDPValtie00tie
ReasoningLlama 3.3 70B Instruct
- GPQA DiamondLlama 3.3 70B Instruct by 940.849.8+9
- CritPttie00tie
- Humanity's Last Examtie3.63.6tie
FactsEven
- AA-Omniscience · Non-hallucinationClaude 3.5 Haiku by 49.158.99.8+49.1
- AA-Omniscience · AccuracyLlama 3.3 70B Instruct by 5.813.218.9+5.8
Long documentsClaude 3.5 Haiku
- AA-LCRClaude 3.5 Haiku by 11.627.315.7+11.6
Following instructionsLlama 3.3 70B Instruct
- IFBenchLlama 3.3 70B Instruct by 4.342.847.1+4.3
Other results14 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.
- AA-OmniscienceClaude 3.5 Haiku by 31.7-22.5-54.2+31.7
- MMLULlama 3.3 70B Instruct by 18.967.186+18.9
- GSM8KLlama 3.3 70B Instruct by 12.781.594.2+12.7
- MATHClaude 3.5 Haiku by 10.287.277+10.2
- naturalquestions_closedbookLlama 3.3 70B Instruct by 8.734.443.1+8.7
- OpenBookQALlama 3.3 70B Instruct by 7.485.492.8+7.4
- DROPLlama 3.3 70B Instruct by 7.183.190.2+7.1
- MGSMLlama 3.3 70B Instruct by 5.585.691.1+5.5
- Artificial Analysis Coding IndexClaude 3.5 Haiku by 415.911.9+4
- MMLU-ProLlama 3.3 70B Instruct by 3.96568.9+3.9
- NarrativeQALlama 3.3 70B Instruct by 2.876.379.1+2.8
- τ²-Bench Telecom (AA run)Llama 3.3 70B Instruct by 224.626.6+2
- AA IntelligenceClaude 3.5 Haiku by 1.28.97.7+1.2
- HumanEvaltie88.188.4tie
Questions people ask
Which is better, Claude 3.5 Haiku or Llama 3.3 70B Instruct?
Claude 3.5 Haiku wins three of the six areas where both have results: coding, agents and long documents. Llama 3.3 70B Instruct wins reasoning and following instructions. They are level on facts.
Which is better for coding?
Claude 3.5 Haiku. It wins 2 of the 3 coding tests both models report; Llama 3.3 70B Instruct wins none, and 1 is a tie.
How do you compare the two?
We use the 26 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 14 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.