Models / Llama 3.3 70B Instruct
Llama 3.3 70B Instruct
Llama 3.3 70B Instruct trails most of the field on what it has been measured on; worth a look for simple, high-volume work where cost matters most. The cheapest model of similar intelligence is Gemma 3 4B, at $0.04 per million tokens against $0.71.
Results 16 Nov 2026 · 28 results · 4 sources
Test Llama 3.3 70B Instruct on your prompt → It runs beside the models you could use instead, with every answer checked and costed.
- Price per million tokens
- $0.71 in · $0.71 out
Typical host on OpenRouter, 10 Nov 2026 · compare prices - Speed
- 98 tokens a second; on its test prompt the answer starts after 0.6 s, thinking time included
Artificial Analysis, on its usual API - Intelligence Index
- 7.7
Artificial Analysis - Spring Prompt overall
- Not ranked yet
needs our benchmarks and two groups of results - Developer
- Meta, based in the US
Weights published - Released
- 6 Dec 2024
131,072 tokens of context
Where it stands among the models you could choose
Available models grouped by price, intelligence, speed and developer location, plus models with published weights. Retired models are excluded.
| Among | Intelligence Index | Spring Prompt overall | Price, blended | Speed |
|---|---|---|---|---|
| Models available today | 152nd of 185Claude Opus 5.5 | – | $0.71 blended, 47th percentilecheapest: Mistral Nemo | 36th of 65Nemotron 3.5 Super VL |
| Models priced $0.25–1 per million tokens | 54th of 61MiMo-V2.6-Pro | – | sets this group | 15th of 20Gemini 3.5 Flash-Lite |
| Models with similar intelligence | sets this group | – | $0.71 blended, 77th percentilecheapest: Gemma 3 4B | 9th of 17gpt-oss-20b |
| Models writing 50–100 tokens a second | 17th of 21Claude Opus 5.5 | – | $0.71 blended, 45th percentilecheapest: Granite 4.2 8B | sets this group |
| Models from developers based in the US | 76th of 92Claude Opus 5.5 | – | $0.71 blended, 36th percentilecheapest: Gemma 3 4B | 22nd of 33Nemotron 3.5 Super VL |
| Open-weights models | 78th of 104MiMo-V2.6-Pro | – | $0.71 blended, 66th percentilecheapest: Mistral Nemo | 21st of 43Nemotron 3.5 Super VL |
The name under each rank is the leader of that group; green is the top quarter of the group and red the bottom quarter, by rank. Price is shown as a percentile: the share of the group that costs less, so lower is cheaper. "Sets this group" marks the measure the group is defined by. Spring Prompt overall is our score out of 100 across every source (how it works). Price is the developer's list price (or the typical OpenRouter host where we haven't read one) for three input tokens to one output; intelligence and speed are from Artificial Analysis (data sourced from Artificial Analysis); "similar intelligence" means within 4 points on its Intelligence Index. Developer locations are where each company is based, not where a model is served. Groups under 3 models are not ranked.
How good is it, and for what?
Each use case ranks the available models its benchmarks measured, then those priced $0.25–1 per million tokens. The bar shows its percentile in that field, best to the right. Open a row for the results behind it.
Product listingsTurning a sparse product feed and photos into listings that can go live
- CatalogBench: has not measured this model.
Decks from an analysisTurning a finished analysis into a deck you could present as it is
- DeckBench: has not measured this model.
User surveysPlanning a user survey and reading its results without being misled
- SurveyBench: has not measured this model.
Marketing planningPlanning a year of ad spend without overspending
- ROASBench: has not measured this model.
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls
| Benchmark | Llama 3.3 70B Instruct | Rank among available | Best available |
|---|---|---|---|
| Berkeley Function Calling Leaderboard (BFCL) V4BFCL: overall accuracy · reported | 31.9% | 24th of 36 | Claude Opus 4.5 77.5% |
| Artificial AnalysisArtificial Analysis: Terminal-Bench 2.1 · reported | 4.9% | 97th of 107 | Claude Fable 5.1 91.4% |
- tau2-bench: has not measured this model.
- OpenHands Index: has not measured this model.
- Microsoft STATE-Bench: has not measured this model.
- Vending-Bench 2: has not measured this model.
Professional workReal tasks from banking, consulting and law, business documents and freelance projects
- APEX-Agents: has not measured this model.
- GDP.pdf: has not measured this model.
- Remote Labor Index: has not measured this model.
Reasoning and knowledgeHard questions across science, maths and general knowledge
| Benchmark | Llama 3.3 70B Instruct | Rank among available | Best available |
|---|---|---|---|
| Artificial AnalysisArtificial Analysis: Artificial Analysis Intelligence Index · reported | 7.7 | 152nd of 185 | Claude Opus 5.5 57.6 |
| Artificial AnalysisArtificial Analysis: Humanity's Last Exam · reported | 3.6% | 177th of 186 | Claude Opus 5.5 61.4% |
| Artificial AnalysisArtificial Analysis: GPQA Diamond · reported | 49.8% | 149th of 168 | GPT-6 Astra 96.3% |
WritingWhat people prefer in blind comparisons, and judged writing quality
| Benchmark | Llama 3.3 70B Instruct | Rank among available | Best available |
|---|---|---|---|
| Arena (formerly LMArena)Arena Text: overall · reported | 1,318 | 131st of 153 | Gemini 4 Argon 1,525 |
| UGI LeaderboardUGI: writing score · reported | 26.2 | 112th of 124 | Gemini 3.8 Flash 78.6 |
Sticking to the factsSummarising without inventing things, and factual answers
- Vectara Hallucination Leaderboard: has not measured this model.
- Arena (formerly LMArena): has not measured this model.
- SimpleQA Verified (Epoch AI): has not measured this model.
Speed under pressureGood decisions against a real clock (fast chess)
- BulletBench: has not measured this model.
Nearest alternatives
- Cheaper at similar intelligenceGemma 3 4B$0.04 blended · Intelligence Index 4.8Compare with Llama 3.3 70B Instruct →
- Meta's newer, at least as capable modelMuse Wonder$8.00 blended · Intelligence Index 52.4
Against Llama 3.3 70B Instruct at $0.71 blended, Intelligence Index 7.7. "Similar intelligence" means within 4 points or better.
Where to run it
9 hosts on OpenRouter at the standard service tier, with a median of $0.71 blended per million tokens; quantisation where the host states it: FP8, FP16. Cheapest first.
| Host | Quantisation | Input | Output | Blended |
|---|---|---|---|---|
| DeepInfra | FP8 | $0.10 | $0.32 | $0.16 |
| AkashML | FP8 | $0.20 | $0.52 | $0.28 |
| Parasail | FP8 | $0.22 | $0.50 | $0.29 |
| Groq | not stated | $0.59 | $0.79 | $0.64 |
| CoreWeave | FP16 | $0.71 | $0.71 | $0.71 |
| not stated | $0.72 | $0.72 | $0.72 | |
| not stated | $0.72 | $0.72 | $0.72 | |
| SambaNova | not stated | $0.60 | $1.20 | $0.75 |
Prices per million tokens from OpenRouter's public endpoint list, checked 10 Nov 2026; the 1 dearer hosts are left out.
Against the alternatives
The models you would most likely weigh it against: the leaders of the groups above.
Scroll sideways to see every alternative.
| Llama 3.3 70B Instruct | MiMo-V2.6-Pro most intelligent at $0.25–1 per million tokens | Gemma 3 4B cheapest with similar intelligence | gpt-oss-20b fastest with similar intelligence | Claude Opus 5.5 most intelligent available today | |
|---|---|---|---|---|---|
| Price per million tokens | $0.71 | $0.54 | $0.04 | $0.06 | $8.00 |
| Tokens a second | 98 | 43 | – | 277 | 96 |
| Intelligence Index | 7.7 | 46.3 | 4.8 | 10.0 | 57.6 |
| Spring Prompt overall | – | – | – | – | 82 |
| Rank among available models, by use case | |||||
| Product listings | – | – | – | – | 13th |
| Decks from an analysis | – | – | – | – | 7th |
| User surveys | – | – | – | – | 4th |
| Marketing planning | – | – | – | – | 9th |
| Agents and tool use | 127th of 159 | 23rd | 158th | 130th | 3rd |
| Professional work | – | – | – | – | 3rd |
| Reasoning and knowledge | 164th of 187 | 13th | 181st | 132nd | 1st |
| Writing | 151st of 169 | 23rd | 157th | 155th | 5th |
| Sticking to the facts | – | 10th | 32nd | – | 3rd |
| Speed under pressure | – | – | – | – | – |
| Side by side → | Side by side → | Side by side → | Side by side → | ||
Shaded figures are better than Llama 3.3 70B Instruct's. A dash means no result.
Research and examples
Where it has been measured
54th most intelligent of the 61 models priced $0.25–1 per million tokens. Among models available today, in the bottom quarter for agents and tool use, reasoning and knowledge, and writing.
- Arena (formerly LMArena)6 results
- Artificial Analysis11 results
- Berkeley Function Calling Leaderboard (BFCL) V48 results
- UGI Leaderboard3 results
- BulletBenchNot measured
- CatalogBenchNot measured
- DeckBenchNot measured
- ROASBenchNot measured
- SurveyBenchNot measured
- APEX-AgentsNot measured
- GDP.pdfNot measured
- Microsoft STATE-BenchNot measured
- OpenHands IndexNot measured
- Remote Labor IndexNot measured
- SimpleQA Verified (Epoch AI)Not measured
- tau2-benchNot measured
- Vectara Hallucination LeaderboardNot measured
- Vending-Bench 2Not measured
Every result
28 published results from 4 sources, each in the source's own units, with the configuration that produced it.
Arena (formerly LMArena) · reported by Arena (formerly LMArena) · 6 results
| Measure | Value | Rank | Configuration | Dated |
|---|---|---|---|---|
| Arena Text: business, management and financeArena rating, higher is better | 1,304 1,296–1,311 |
266 of 406 | Llama 3.3 70B Instruct | 2 Oct 2026 |
| Arena Text: creative writingArena rating, higher is better | 1,285 1,278–1,292 |
254 of 411 | Llama 3.3 70B Instruct | 2 Oct 2026 |
| Arena Text: expert promptsArena rating, higher is better | 1,298 1,286–1,309 |
262 of 363 | Llama 3.3 70B Instruct | 2 Oct 2026 |
| Arena Text: instruction followingArena rating, higher is better | 1,293 1,288–1,298 |
270 of 413 | Llama 3.3 70B Instruct | 2 Oct 2026 |
| Arena Text: overallArena rating, higher is better | 1,318 1,314–1,321 |
261 of 413 | Llama 3.3 70B Instruct | 2 Oct 2026 |
| Arena Text: writing, literature and languageArena rating, higher is better | 1,292 1,286–1,297 |
268 of 412 | Llama 3.3 70B Instruct | 2 Oct 2026 |
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.
Artificial Analysis · reported by Artificial Analysis · 11 results
| Measure | Value | Rank | Configuration | Dated |
|---|---|---|---|---|
| Artificial Analysis: Artificial Analysis Coding Indexindex score, higher is better | 11.9 | 172 of 187 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better | 7.7 | 357 of 413 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: GPQA Diamond% of questions, higher is better | 49.8% | 323 of 360 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: Humanity's Last Exam% of questions, higher is better | 3.6% | 401 of 419 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: IFBench% of instructions, higher is better | 47.1% | 156 of 288 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better | 15.7% | 376 of 407 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: Output speedtokens per second, higher is better | 97.7 | 69 of 137 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: Terminal-Bench 2.1% of tasks, higher is better | 4.9% | 168 of 188 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better | 3.0% | 255 of 282 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: Time to first answer tokenseconds, lower is better | 0.64 | 14 of 137 | Llama 3.3 70B Instruct | 20 Oct 2026 |
| Artificial Analysis: τ²-bench telecom% of tasks, higher is better | 26.6% | 227 of 286 | Llama 3.3 70B Instruct | 20 Oct 2026 |
Not shown: Not your codebase or tools.
Berkeley Function Calling Leaderboard (BFCL) V4 · reported by Berkeley Function Calling Leaderboard (BFCL) V4 · 8 results
| Measure | Value | Rank | Configuration | Dated |
|---|---|---|---|---|
| BFCL: irrelevance detection% correct, higher is better | 53.5% | 97 of 109 | Llama 3.3 70B Instruct (bfcl fc) | 16 Dec 2025 |
| BFCL: memory% correct, higher is better | 8.2% | 79 of 109 | Llama 3.3 70B Instruct (bfcl fc) | 16 Dec 2025 |
| BFCL: multi-turn tasks% correct, higher is better | 21.5% | 50 of 109 | Llama 3.3 70B Instruct (bfcl fc) | 16 Dec 2025 |
| BFCL: overall accuracy% correct, higher is better | 31.9% | 62 of 109 | Llama 3.3 70B Instruct (bfcl fc) | 16 Dec 2025 |
| BFCL: relevance detection% correct, higher is better | 100.0% | 1 of 109 | Llama 3.3 70B Instruct (bfcl fc) | 16 Dec 2025 |
| BFCL: single-turn calls (curated)% correct, higher is better | 88.0% | 22 of 109 | Llama 3.3 70B Instruct (bfcl fc) | 16 Dec 2025 |
| BFCL: single-turn calls (user-contributed)% correct, higher is better | 76.6% | 33 of 109 | Llama 3.3 70B Instruct (bfcl fc) | 16 Dec 2025 |
| BFCL: web search% correct, higher is better | 10.0% | 56 of 109 | Llama 3.3 70B Instruct (bfcl fc) | 16 Dec 2025 |
Not shown: Not general refusal or safety behaviour.
UGI Leaderboard · reported by UGI Leaderboard · 3 results
| Measure | Value | Rank | Configuration | Dated |
|---|---|---|---|---|
| UGI: requested-length error% off the requested word count, lower is better | 10.0% | 89 of 379 | Llama 3.3 70B Instruct | 10 Sep 2025 |
| UGI: style adherencescore from 0 to 1, higher is better | 0.37 | 110 of 379 | Llama 3.3 70B Instruct | 10 Sep 2025 |
| UGI: writing scorescore out of 100, higher is better | 26.2 | 304 of 379 | Llama 3.3 70B Instruct | 10 Sep 2025 |
Not shown: Not other format limits such as character counts or bullet counts.
Will Llama 3.3 70B Instruct work for your product?
Benchmarks aren't your task. Paste the prompt your product uses and run it beside the models you could use instead, with every answer checked and costed.
New here? Your email starts a free trial. No credit card needed.