Compare · ModelsLive · 2 picked · head to head
Llama 3.1 70B Instruct vs Llama 3.2 90B
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Llama 3.1 70B Instruct wins on 3/5 benchmarks
Llama 3.1 70B Instruct wins 3 of 5 shared benchmarks. Leads in knowledge.
Category leads
knowledge·Llama 3.1 70B Instructmath·Llama 3.2 90B
Hype vs Reality
Attention vs performance
Llama 3.1 70B Instruct
#152 by perf·no signal
Llama 3.2 90B
#162 by perf·no signal
Best value
Llama 3.1 70B Instruct
Llama 3.1 70B Instruct
94.5 pts/$
$0.40/M
Llama 3.2 90B
—
no price
Vendor risk
Who is behind the model
Meta AI
$1.50T·Tier 1
Meta AI
$1.50T·Tier 1
Head to head
5 benchmarks · 2 models
Llama 3.1 70B InstructLlama 3.2 90B
Balrog
Llama 3.1 70B Instruct leads by +0.6
Balrog · benchmarks AI agents on text-based adventure games, testing language understanding, strategic planning, and long-horizon reasoning.
Llama 3.1 70B Instruct
27.9
Llama 3.2 90B
27.3
GPQA diamond
Llama 3.1 70B Instruct leads by +4.2
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Llama 3.1 70B Instruct
25.6
Llama 3.2 90B
21.4
MATH level 5
Llama 3.2 90B leads by +2.8
MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics.
Llama 3.1 70B Instruct
36.7
Llama 3.2 90B
39.4
MMLU
Llama 3.2 90B leads by +0.3
Massive Multitask Language Understanding · 57 subjects spanning STEM, humanities, social sciences, and more. The standard benchmark for broad knowledge.
Llama 3.1 70B Instruct
73.5
Llama 3.2 90B
73.7
OTIS Mock AIME 2024-2025
Llama 3.1 70B Instruct leads by +1.0
OTIS Mock AIME 2024–2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Llama 3.1 70B Instruct
3.5
Llama 3.2 90B
2.5
Full benchmark table
| Benchmark | Llama 3.1 70B Instruct | Llama 3.2 90B |
|---|---|---|
Balrog Balrog · benchmarks AI agents on text-based adventure games, testing language understanding, strategic planning, and long-horizon reasoning. | 27.9 | 27.3 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 25.6 | 21.4 |
MATH level 5 MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics. | 36.7 | 39.4 |
MMLU Massive Multitask Language Understanding · 57 subjects spanning STEM, humanities, social sciences, and more. The standard benchmark for broad knowledge. | 73.5 | 73.7 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024–2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 3.5 | 2.5 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| Llama 3.1 70B Instruct | $0.40 | $0.40 | 131K tokens (~66 books) | $4.00 |
| Llama 3.2 90B | — | — | — | — |