Claude 3.5 Haiku vs Llama 3.1 8B Instruct
Wins 5 of 6 areas
Coding · Agents · Facts · Long documents · Following instructions
Wins 0 of 6 areas
—
Claude 3.5 Haiku is the stronger all-rounder.Llama 3.1 8B Instruct is cheaper.
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 2 tests
- FactsGetting facts right instead of making them up20Claude 3.5 Haiku2 of 2 tests
- 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
- Following instructionsDoing exactly what it is asked10Claude 3.5 Haiku1 of 1 test
- ReasoningHard problems that need careful thinking11Even1 each · 1 tie
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.
Llama 3.1 8B Instruct costs 99% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Claude 3.5 Haiku pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+14.9points ahead
- Code for real scientific research problemsSciCode+14.2points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+14.2points ahead
Where Llama 3.1 8B Instruct pulls ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+1.7points 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
- SciCodeClaude 3.5 Haiku by 14.227.413.2+14.2
- Terminal-Bench HardClaude 3.5 Haiku by 1.52.30.8+1.5
AgentsClaude 3.5 Haiku
- τ-Bench V3 · BankingClaude 3.5 Haiku by 4.25.81.6+4.2
- GDPValtie00tie
ReasoningEven
- GPQA DiamondClaude 3.5 Haiku by 14.940.825.9+14.9
- Humanity's Last ExamLlama 3.1 8B Instruct by 1.73.65.3+1.7
- CritPttie00tie
FactsClaude 3.5 Haiku
- AA-Omniscience · AccuracyClaude 3.5 Haiku by 4.713.28.5+4.7
- AA-Omniscience · Non-hallucinationClaude 3.5 Haiku by 1.958.957+1.9
Following instructionsClaude 3.5 Haiku
- IFBenchClaude 3.5 Haiku by 14.242.828.6+14.2
Other results15 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.
- MATHClaude 3.5 Haiku by 35.387.251.9+35.3
- DROPClaude 3.5 Haiku by 23.683.159.5+23.6
- MGSMClaude 3.5 Haiku by 16.785.668.9+16.7
- MMLU-ProClaude 3.5 Haiku by 16.76548.3+16.7
- HumanEvalClaude 3.5 Haiku by 15.588.172.6+15.5
- MBPPClaude 3.5 Haiku by 15.585.670.1+15.5
- naturalquestions_closedbookClaude 3.5 Haiku by 13.534.420.9+13.5
- OpenBookQAClaude 3.5 Haiku by 11.485.474+11.4
- Artificial Analysis Coding IndexClaude 3.5 Haiku by 10.515.95.4+10.5
- AA-OmniscienceClaude 3.5 Haiku by 8.4-22.5-30.9+8.4
- τ²-Bench Telecom (AA run)Claude 3.5 Haiku by 8.224.616.4+8.2
- MMLULlama 3.1 8B Instruct by 5.967.173+5.9
- AA IntelligenceClaude 3.5 Haiku by 28.96.9+2
- GSM8KClaude 3.5 Haiku by 1.781.579.8+1.7
- NarrativeQAtie76.375.6tie
Questions people ask
Which is better, Claude 3.5 Haiku or Llama 3.1 8B Instruct?
Claude 3.5 Haiku wins five of the six areas where both have results: coding, agents, facts, long documents and following instructions. Llama 3.1 8B Instruct wins none, but costs 99% less. They are level on reasoning.
Which is better for coding?
Claude 3.5 Haiku. It wins 2 of the 2 coding tests both models report; Llama 3.1 8B Instruct wins none.
Which is cheaper?
Claude 3.5 Haiku costs $0.80 per million input tokens and $4.00 per million output tokens; Llama 3.1 8B Instruct costs $0.02 and $0.05. That makes Llama 3.1 8B Instruct about 99% cheaper for the same work.
How do you compare the two?
We use the 26 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 15 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.