DeepSeek-V3 vs Llama 3.1 70B Instruct
Wins 4 of 6 areas
Coding · Reasoning · Long documents · Following instructions
Wins 0 of 6 areas
—
DeepSeek-V3 is the stronger all-rounder.Llama 3.1 70B Instruct is cheaper.
Scores updated · 35 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 software20DeepSeek-V32 of 2 tests
- ReasoningHard problems that need careful thinking10DeepSeek-V31 of 3 tests · 2 ties
- Long documentsFinding answers in very long texts10DeepSeek-V31 of 1 test
- Following instructionsDoing exactly what it is asked10DeepSeek-V31 of 1 test
- 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.
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 70B Instruct costs 2% 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 DeepSeek-V3 pulls ahead
- Reasons across sets of long documentsAA-LCR+32.7points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+24.6points ahead
- Code for real scientific research problemsSciCode+12.3points ahead
Where Llama 3.1 70B Instruct pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+7.7points ahead
Every test, side by side
All 35 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V3
- SciCodeDeepSeek-V3 by 12.33926.7+12.3
- Terminal-Bench HardDeepSeek-V3 by 12.215.23+12.2
ReasoningDeepSeek-V3
- GPQA DiamondDeepSeek-V3 by 24.665.540.9+24.6
- Humanity's Last Examtie4.74.5tie
- CritPttie00tie
FactsEven
- AA-Omniscience · Non-hallucinationLlama 3.1 70B Instruct by 7.714.121.8+7.7
- AA-Omniscience · AccuracyDeepSeek-V3 by 5.825.419.7+5.8
Following instructionsDeepSeek-V3
- IFBenchDeepSeek-V3 by 6.64134.4+6.6
Other results25 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.
- Arena HardDeepSeek-V3 by 35.791.455.7+35.7
- τ²-Bench Telecom (AA run)DeepSeek-V3 by 31.947.115.2+31.9
- MATHDeepSeek-V3 by 22.290.268+22.2
- HumanEvalLlama 3.1 70B Instruct by 15.365.280.5+15.3
- Artificial Analysis Coding IndexDeepSeek-V3 by 12.12310.9+12.1
- DROPDeepSeek-V3 by 1291.679.6+12
- MMLU-ProDeepSeek-V3 by 9.575.966.4+9.5
- MGSMLlama 3.1 70B Instruct by 7.179.886.9+7.1
- MBPPDeepSeek-V3 by 6.875.468.6+6.8
- BBHDeepSeek-V3 by 5.987.581.6+5.9
- anthropic_red_teamDeepSeek-V3 by 3.997.193.2+3.9
- AA IntelligenceDeepSeek-V3 by 3.19.76.6+3.1
- HarmBenchDeepSeek-V3 by 2.849.746.9+2.8
- simple_safety_testsDeepSeek-V3 by 2.895.392.5+2.8
- XSTestDeepSeek-V3 by 2.697.194.5+2.6
- MMLUDeepSeek-V3 by 2.588.586+2.5
- AA-OmniscienceDeepSeek-V3 by 2.4-40.7-43.1+2.4
- NarrativeQADeepSeek-V3 by 2.479.677.2+2.4
- AIR-Bench 2024Llama 3.1 70B Instruct by 1.740.842.5+1.7
- OpenBookQADeepSeek-V3 by 1.695.493.8+1.6
- naturalquestions_closedbookDeepSeek-V3 by 1.546.745.2+1.5
- IFEvalLlama 3.1 70B Instruct by 1.486.187.5+1.4
- ARC-Challengetie95.394.8tie
- WinoGrandetie84.985.3tie
- GSM8Ktie9493.8tie
Questions people ask
Which is better, DeepSeek-V3 or Llama 3.1 70B Instruct?
DeepSeek-V3 wins four of the six areas where both have results: coding, reasoning, long documents and following instructions. Llama 3.1 70B Instruct wins none, but costs 2% less. They are level on agents and facts.
Which is better for coding?
DeepSeek-V3. It wins 2 of the 2 coding tests both models report; Llama 3.1 70B Instruct wins none.
Which is cheaper?
DeepSeek-V3 costs $0.24 per million input tokens and $0.90 per million output tokens; Llama 3.1 70B Instruct costs $0.56 and $0.56. That makes Llama 3.1 70B Instruct about 2% cheaper for the same work.
How do you compare the two?
We use the 35 benchmark tests both models have published scores on. The verdict counts the 10 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 25 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.