# Meta Llama 3.3 70B Instruct

As of October 22, 2026, Meta Llama 3.3 70B Instruct ranks #115 of 123 models overall with 26.0 ±11.0 on the LLMBenchmark Index. Llama 3.3 70B Instruct ranks #105 of 122 in reasoning, its strongest category. Meta Llama 3.3 70B Instruct costs $0.45 per million input tokens and $0.72 per million output tokens. Llama 3.3 70B Instruct was released on December 6, 2024.

Page: https://llmbenchmark.ai/models/llama-3-3-70b-instruct
Data from October 22, 2026. Every figure below names the board it was read from and the day.

- Maker: Meta Llama
- Released: December 6, 2024
- Price per million input tokens: $0.45
- Price per million output tokens: $0.72
- Context window: 131K tokens
- Overall rank: 115 of 123, by the LLMBenchmark Index
- LLMBenchmark Index: 26.0 ±11.0 (90% margin, from 4 of 7 inputs)
- Artificial Analysis Intelligence Index: 7.7

Benchmark results (52 of 195):

- AA Intelligence Index: 7.7 (source: Artificial Analysis (via OpenRouter), read October 22, 2026)
- AA Coding Index: 11.9 (source: Artificial Analysis (via OpenRouter), read October 22, 2026)
- AA Math Index: 7.7 (source: Artificial Analysis, read October 22, 2026)
- GPQA Diamond (Artificial Analysis run): 49.8% (source: Artificial Analysis, read October 22, 2026)
- Humanity's Last Exam (Artificial Analysis run): 3.6% (source: Artificial Analysis, read October 22, 2026)
- MMLU Pro (Artificial Analysis run): 71.3% (source: Artificial Analysis, read October 22, 2026)
- LiveCodeBench (Artificial Analysis run): 28.8% (source: Artificial Analysis, read October 22, 2026)
- Terminal-Bench Hard (Artificial Analysis run): 3.0% (source: Artificial Analysis, read October 22, 2026)
- IFBench: 47.1% (source: Artificial Analysis, read October 22, 2026)
- Long context reasoning (AA-LCR): 15.7% (source: Artificial Analysis, read October 22, 2026)
- τ²-Bench Telecom: 26.6% (source: Artificial Analysis, read October 22, 2026)
- AIME 2025 (Artificial Analysis run): 7.7% (source: Artificial Analysis, read October 22, 2026)
- WeirdML: 14.4% (source: Epoch AI benchmarking hub, read October 22, 2026)
- BALROG: 23.0% (source: Epoch AI benchmarking hub, read October 22, 2026)
- GPQA Diamond (Epoch AI run): 47.4% (source: Epoch AI benchmarking hub, read October 22, 2026)
- SimpleBench: 19.9% (source: Epoch AI benchmarking hub, read October 22, 2026)
- OTIS Mock AIME: 5.1% (source: Epoch AI benchmarking hub, read October 22, 2026)
- MATH level 5: 41.6% (source: Epoch AI benchmarking hub, read October 22, 2026)
- MMLU: 86.3% (source: Epoch AI benchmarking hub, read October 22, 2026)
- DTBench: 59.5% (source: Epoch AI benchmarking hub, read October 22, 2026)
- LMCA: 17.5% (source: Epoch AI benchmarking hub, read October 22, 2026)
- Fiction.LiveBench: 33.3% (source: Epoch AI benchmarking hub, read October 22, 2026)
- Epoch Capabilities Index: 127.3 (source: Epoch AI benchmarking hub, read October 22, 2026)
- LMArena Text: 1274 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: coding: 1268 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: hard prompts: 1257 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: math: 1267 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: instruction following: 1242 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: creative writing: 1250 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: English: 1303 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: non-English: 1236 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: Chinese: 1217 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: German: 1251 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: Japanese: 1150 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: Korean: 1142 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: French: 1282 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: Spanish: 1269 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: Russian: 1252 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: Polish: 1259 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: multi-turn: 1280 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: longer queries: 1255 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: expert: 1225 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: hard prompts in English: 1277 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: software and IT: 1273 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: mathematical work: 1267 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: science: 1284 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: medicine and healthcare: 1275 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: legal and government: 1284 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: business and finance: 1243 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: writing and language: 1255 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- LMArena Text: entertainment and media: 1241 Elo (source: LMArena (Hugging Face dataset), read October 22, 2026)
- Berkeley Function Calling Leaderboard (BFCL): 31.9% (source: Berkeley Function Calling Leaderboard, read October 22, 2026)