Llama 3.3 70B Instruct vs Mistral Large 2
Wins 2 of 6 areas
Long documents · Following instructions
Wins 2 of 6 areas
Coding · Facts
The two are evenly matched.Llama 3.3 70B Instruct is better at long documents and following instructions; Mistral Large 2 at coding and facts.
Scores updated · 22 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 software02Mistral Large 22 of 2 tests
- FactsGetting facts right instead of making them up01Mistral Large 21 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts10Llama 3.3 70B Instruct1 of 1 test
- Following instructionsDoing exactly what it is asked10Llama 3.3 70B Instruct1 of 1 test
- AgentsCarrying out multi-step tasks on its own00Even0 each · 1 tie
- ReasoningHard problems that need careful thinking11Even1 each · 2 ties
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 3.3 70B Instruct costs 82% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Mistral Large 2 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+22.5points ahead
- Code for real scientific research problemsSciCode+3.2points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+3.1points ahead
Where Llama 3.3 70B Instruct pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+15.9points ahead
- Reasons across sets of long documentsAA-LCR+13.7points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+1.2points ahead
Every test, side by side
All 22 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingMistral Large 2
- SciCodeMistral Large 2 by 3.22629.2+3.2
- Terminal-Bench HardMistral Large 2 by 3.136.1+3.1
ReasoningEven
- SimpleBenchMistral Large 2 by 2.619.922.5+2.6
- GPQA DiamondLlama 3.3 70B Instruct by 1.249.848.6+1.2
- Humanity's Last Examtie3.63.3tie
- CritPttie00tie
FactsMistral Large 2
- AA-Omniscience · Non-hallucinationMistral Large 2 by 22.59.832.3+22.5
- AA-Omniscience · Accuracytie18.919.9tie
Long documentsLlama 3.3 70B Instruct
- AA-LCRLlama 3.3 70B Instruct by 13.715.72+13.7
Following instructionsLlama 3.3 70B Instruct
- IFBenchLlama 3.3 70B Instruct by 15.947.131.2+15.9
Other results11 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-OmniscienceMistral Large 2 by 19.8-54.2-34.4+19.8
- MMLULlama 3.3 70B Instruct by 13.58672.5+13.5
- MATHLlama 3.3 70B Instruct by 9.37767.7+9.3
- τ²-Bench Telecom (AA run)Mistral Large 2 by 4.126.630.7+4.1
- HumanEvalMistral Large 2 by 3.688.492+3.6
- naturalquestions_closedbookMistral Large 2 by 2.243.145.3+2.2
- Artificial Analysis Coding IndexMistral Large 2 by 1.911.913.8+1.9
- GSM8KLlama 3.3 70B Instruct by 1.294.293+1.2
- NarrativeQALlama 3.3 70B Instruct by 1.279.177.9+1.2
- OpenBookQAtie92.893.2tie
- AA Intelligencetie7.77.6tie
Questions people ask
Which is better, Llama 3.3 70B Instruct or Mistral Large 2?
Llama 3.3 70B Instruct and Mistral Large 2 each win two of the six areas where both have results. Llama 3.3 70B Instruct wins long documents and following instructions; Mistral Large 2 wins coding and facts. Llama 3.3 70B Instruct costs 82% less. They are level on agents and reasoning.
Which is better for coding?
Mistral Large 2. It wins 2 of the 2 coding tests both models report; Llama 3.3 70B Instruct wins none.
Which is cheaper?
Llama 3.3 70B Instruct costs $0.71 per million input tokens and $0.72 per million output tokens; Mistral Large 2 costs $2.00 and $6.00. That makes Llama 3.3 70B Instruct about 82% cheaper for the same work.
How do you compare the two?
We use the 22 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 11 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.