MetaLlama 3.3 70B
Llama 3.3 70B is a Meta open-weights model, released 6 Dec 2024. Its best configuration ranks #651 of 785 on the BenchLeader Index at 40.7 ±4.1, in the lower half. The ± is the point: 273 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (49) and lowest in coding (31). At $0.675 per million tokens blended it is mid-priced. Output speed of 92 tokens per second puts it faster than most, with a first answer in 1.8 s. Last measured 22 Oct 2026.
- Blended price
- $0.675/M
- $0.660 in · $0.720 out
- Output speed
- 92 tok/s
- measured by Artificial Analysis
- First answer
- 1.76 s
- first token 0.52 s
- Context
- 128k
- Full answer
- 7.20 s
- median, reasoning included
- Cost per run
- –
- Released
- 6 Dec 2024
- Overall index41
- Reasoning36
- Coding31
- Agents & tools42
- Maths40
- Knowledge37
- Instruction following49
- Human preference47
- Long context32
- Composite38
Versions
Meta has shipped 4 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| Llama 3.3 70Bthis page | 6 Dec 2024 | 40.7 | #651 |
| Llama 3.1 70B | 23 Jul 2024 | 40.9 | #645 |
| Llama 3-70B | 18 Apr 2024 | 38.6 | #707 |
| Llama 2-70B | 18 Jul 2023 | 34.3 | #772 |
How its index has moved
8 Sept 2026 to 23 Oct 2026The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| GPQA Diamond | 47.4%#227 | Epoch AI Benchmarking Hub | |
| SimpleBench | 19.9%#81 | SimpleBench | |
| Arena Hard Prompts | 1321#241 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 49.8%#431 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 3.6%#589 | Artificial Analysis | |
| CritPt | 0.0%#315 | Artificial Analysis | |
| SciPredictnot in index | 18.2%#9 | Scale AI SEAL | |
| LMCAnot in index | 17.5%#164 | Epoch AI Benchmarking Hub | |
| DTBenchnot in index | 59.5%#189 | Epoch AI Benchmarking Hub | |
| ForecastBenchnot in index | 61.3%#66 | ForecastBench |
Coding
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| SciCode | 26.0%#205 | SciCode | |
| WeirdML | 14.4%#146 | WeirdML | |
| Arena Coding | 1346#247 | Arena (LMArena) | |
| Terminal-Bench 2.1 (AA)not in index | 4.9%#206 | Artificial Analysis |
Agents & tools
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Terminal-Bench Hard | 3.0%#314 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 26.6%#280 | Artificial Analysis | |
| BFCL Overall | 31.9%#46 | Berkeley Function Calling Leaderboard | |
| GDPval-AA v2.1not in index | 0.0%#124 | Artificial Analysis | |
| ITBench SRE (AA)not in index | 0.6%#45 | Artificial Analysis | |
| BALROGnot in index | 23.0%#22 | BALROG |
Maths
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| OTIS Mock AIME | 5.1%#245 | Epoch AI Benchmarking Hub | |
| MATH Level 5 | 41.6%#61 | Epoch AI Benchmarking Hub | |
| Arena Maths | 1295#233 | Arena (LMArena) |
Knowledge
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-Omniscience | -54.2#426 | Artificial Analysis | |
| MMLUnot in index | 86.3%#6 | Epoch AI Benchmarking Hub | |
| AA-Omniscience: accuracynot in index | 18.9%#323 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 9.8%#426 | Artificial Analysis |
Instruction following
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| IFBench | 47.1%#191 | Artificial Analysis | |
| Arena Instruction Followingnot in index | 1293#245 | Arena (LMArena) |
Human preference
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Text | 1318#237 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1285#232 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1317#229 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1314#246 | Arena (LMArena) |
Long context
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-LCR | 15.7%#452 | Artificial Analysis |
Composite
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | 127.3#155 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 7.7#462 | Artificial Analysis |
Safety & honesty
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| FORTRESSnot in index | 44.8%#16 | Scale AI SEAL | |
| MASKnot in index | 51.9%#47 | Scale AI SEAL |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Groq | 228 tok/s | 0.42 s | $0.590 | $0.790 | 131k | – |
| SambaNova Turbo | 87 tok/s | 0.62 s | $0.450 | $0.900 | 131k | – |
| CoreWeave | 80 tok/s | 0.23 s | $0.710 | $0.710 | 128k | fp16 |
| Parasail | 42 tok/s | 0.68 s | $0.220 | $0.500 | 131k | fp8 |
| Cloudflare | 40 tok/s | 0.60 s | $0.293 | $2.25 | 24k | fp8 |
| NovitaAI | 37 tok/s | 0.70 s | $0.135 | $0.400 | 12k | bf16 |
| AkashML | 34 tok/s | 0.65 s | $0.200 | $0.520 | 131k | fp8 |
| DeepInfra (Turbo) | 14 tok/s | 0.58 s | $0.100 | $0.320 | 131k | fp8 |
Price history
The price above, per 1M tokens, every day it changed.
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0005 | – | 5.0 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0084 | – | 8.3 s |
| Code edit | 6,000 / 1,500 | $0.0050 | – | 18.1 s |
| Agentic coding session | 60,000 / 4,000 | $0.043 | – | 45.3 s |
| Structured extraction | 2,000 / 200 | $0.0015 | – | 3.9 s |
More from Meta
See also
Questions people ask
- What are Llama 3.3 70B's benchmark scores?
- As of 24 Oct 2026, Llama 3.3 70B scores 40.7 ±4.1 on the BenchLeader Index, strongest in instruction following (49) and weakest in coding (31). That rests on 17 independent benchmark results, and every score on this page links to its publisher.
- How does Llama 3.3 70B rank?
- Llama 3.3 70B is #651 of 785 ranked configurations on 24 Oct 2026. 273 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does Llama 3.3 70B cost per 1M tokens?
- As of 24 Oct 2026, Llama 3.3 70B costs $0.660 per 1M input tokens and $0.720 per 1M output tokens, $0.675 blended (three input tokens per output token). The list price last changed on 24 Oct 2026.
- When was Llama 3.3 70B released?
- Meta released Llama 3.3 70B on 6 Dec 2024.
Data as of 24 Oct 2026. Compare with another model.
Cite as: BenchLeader, “Llama 3.3 70B: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/llama-3-3-70b.