Llama 3.1 70B Instruct vs Nova Micro
Wins 3 of 6 areas
Coding · Reasoning · Following instructions
Wins 1 of 6 areas
Long documents
Llama 3.1 70B Instruct wins more areas, narrowly.Nova Micro is cheaper and better at long documents.
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 software20Llama 3.1 70B Instruct2 of 2 tests
- ReasoningHard problems that need careful thinking10Llama 3.1 70B Instruct1 of 3 tests · 2 ties
- Following instructionsDoing exactly what it is asked10Llama 3.1 70B Instruct1 of 1 test
- Long documentsFinding answers in very long texts01Nova Micro1 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.
Nova Micro costs 84% 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 Llama 3.1 70B Instruct pulls ahead
- Code for real scientific research problemsSciCode+17.3points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+9.7points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+5.1points ahead
Where Nova Micro pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+13.1points ahead
- Reasons across sets of long documentsAA-LCR+5points 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.
CodingLlama 3.1 70B Instruct
- SciCodeLlama 3.1 70B Instruct by 17.326.79.4+17.3
- Terminal-Bench HardLlama 3.1 70B Instruct by 1.531.5+1.5
ReasoningLlama 3.1 70B Instruct
- GPQA DiamondLlama 3.1 70B Instruct by 5.140.935.8+5.1
- Humanity's Last Examtie4.54.6tie
- CritPttie00tie
FactsEven
- AA-Omniscience · Non-hallucinationNova Micro by 13.121.834.9+13.1
- AA-Omniscience · AccuracyLlama 3.1 70B Instruct by 9.719.710+9.7
Following instructionsLlama 3.1 70B Instruct
- IFBenchLlama 3.1 70B Instruct by 534.429.4+5
Other results16 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.
- BFCLv4Llama 3.1 70B Instruct by 28.684.856.2+28.6
- MMLULlama 3.1 70B Instruct by 228664+22
- naturalquestions_closedbookLlama 3.1 70B Instruct by 16.745.228.5+16.7
- Artificial Analysis Coding IndexLlama 3.1 70B Instruct by 6.810.94.1+6.8
- AA-OmniscienceLlama 3.1 70B Instruct by 5.4-43.1-48.5+5.4
- OpenBookQALlama 3.1 70B Instruct by 593.888.8+5
- ARC-ChallengeLlama 3.1 70B Instruct by 4.694.890.2+4.6
- NarrativeQALlama 3.1 70B Instruct by 2.877.274.4+2.8
- BBHLlama 3.1 70B Instruct by 2.181.679.5+2.1
- GSM8KLlama 3.1 70B Instruct by 1.593.892.3+1.5
- τ²-Bench Telecom (AA run)Llama 3.1 70B Instruct by 1.215.214+1.2
- AA Intelligencetie6.65.9tie
- HumanEvaltie80.581.1tie
- AA Agentic Indextie5.14.7tie
- DROPtie79.679.3tie
- IFEvaltie87.587.2tie
Questions people ask
Which is better, Llama 3.1 70B Instruct or Nova Micro?
Llama 3.1 70B Instruct wins three of the six areas where both have results: coding, reasoning and following instructions. Nova Micro wins long documents, and costs 84% less. They are level on agents and facts.
Which is better for coding?
Llama 3.1 70B Instruct. It wins 2 of the 2 coding tests both models report; Nova Micro wins none.
Which is cheaper?
Llama 3.1 70B Instruct costs $0.56 per million input tokens and $0.56 per million output tokens; Nova Micro costs $0.04 and $0.14. That makes Nova Micro about 84% 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 10 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 16 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.