DeepSeek-V3 is the stronger all-rounder.Llama 4 Maverick 17B 128E Instruct FP8 is cheaper.
Scores updated · 10 tests both models report · How we compare
Where each one wins
Tests won in each of the one area 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.
Facts rests on a single test.
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 4 Maverick 17B 128E Instruct FP8 costs 2% less for the same work.
Every test, side by side
All 10 tests both models report. The winning score is in its model's colour; marks a score checked independently.
FactsDeepSeek-V3
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3 by 2.16.18.2+2.1
Other results9 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.
- AIR-Bench 2024Llama 4 Maverick 17B 128E Instruct FP8 by 27.840.868.6+27.8
- HarmBenchLlama 4 Maverick 17B 128E Instruct FP8 by 16.449.766.1+16.4
- simple_safety_testsLlama 4 Maverick 17B 128E Instruct FP8 by 495.399.3+4
- bbqDeepSeek-V3 by 3.796.793+3.7
- vectara_answer_rateLlama 4 Maverick 17B 128E Instruct FP8 by 2.597.5100+2.5
- vectara_avg_summary_lengthLlama 4 Maverick 17B 128E Instruct FP8 by 24.3 rating points81.7106+24.3 rating
- vectara_factual_consistencyDeepSeek-V3 by 2.193.991.8+2.1
- anthropic_red_teamLlama 4 Maverick 17B 128E Instruct FP8 by 1.197.198.2+1.1
- XSTesttie97.196.5tie
Questions people ask
Which is better, DeepSeek-V3 or Llama 4 Maverick 17B 128E Instruct FP8?
DeepSeek-V3 wins the one area where both have results: facts. Llama 4 Maverick 17B 128E Instruct FP8 wins none, but costs 2% less.
Which is cheaper?
DeepSeek-V3 costs $0.24 per million input tokens and $0.90 per million output tokens; Llama 4 Maverick 17B 128E Instruct FP8 costs $0.27 and $0.85. That makes Llama 4 Maverick 17B 128E Instruct FP8 about 2% cheaper for the same work.
How do you compare the two?
We use the 10 benchmark tests both models have published scores on. The verdict counts the 1 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 9 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.