Compare / Llama 3.3 70B Instruct vs gpt-oss-120b
Llama 3.3 70B Instruct vs gpt-oss-120b
Which is cheaper, which is more capable, and which does better at each kind of work, from 20 results on 3 sources that measured both models.
Llama 3.3 70B Instruct or gpt-oss-120b? The short answer
- gpt-oss-120b is cheaper: $0.26 against $0.64 per million tokens, 2.4 times less.
- gpt-oss-120b scores higher on the Artificial Analysis Intelligence Index: 11.6 against 7.7.
- gpt-oss-120b writes faster: 200 tokens a second against 80.
- On our business benchmarks, gpt-oss-120b ranks higher for agents and tool use, reasoning and knowledge and writing.
At a glance
| Llama 3.3 70B Instruct | gpt-oss-120b | |
|---|---|---|
| Price per million tokens | $0.59 in · $0.79 out | $0.15 in · $0.60 out |
| Blended (3 in : 1 out) | $0.64 | $0.26 |
| Cheapest host, blended | $0.16 (DeepInfra, FP8) | $0.065 (CoreWeave, FP4) |
| Intelligence Index | 7.7 | 11.6 |
| Spring Prompt overall | – | – |
| Output speed, tokens a second | 80 | 200 |
| Context window | 131,072 tokens | 131,072 tokens |
| Weights | Open | Open |
| Developer based in | the US | the US |
Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.
Which is better for what
Each model's rank among models available today, for each kind of work, from our own benchmarks and licensed sources. The better rank is in green.
| Use case | Llama 3.3 70B Instruct | gpt-oss-120b |
|---|---|---|
| Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls | 127 of 161 | 117 of 161 |
| Professional workReal tasks from banking, consulting and law, business documents and freelance projects | not measured | 49 of 51 |
| Reasoning and knowledgeHard questions across science, maths and general knowledge | 174 of 195 | 110 of 195 |
| WritingWhat people prefer in blind comparisons, and judged writing quality | 165 of 187 | 131 of 187 |
| Sticking to the factsSummarising without inventing things, and factual answers | not measured | 138 of 153 |
Every shared result
| Benchmark | Metric | Llama 3.3 70B Instruct | gpt-oss-120b |
|---|---|---|---|
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,304 | 1,351 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,285 | 1,277 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,298 | 1,357 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,293 | 1,323 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,318 | 1,352 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,292 | 1,309 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 11.9 | 30.4 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 7.70 | 11.6 |
| Artificial Analysisreported | GPQA Diamond% of questions | 49.8% | 78.2% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 3.6% | 19.6% |
| Artificial Analysisreported | IFBench% of instructions | 47.1% | 69.0% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 15.7% | 52.0% |
| Artificial Analysisreported | Output speedtokens per second | 80.5 | 200 |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 4.9% | 26.2% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 3.0% | 23.5% |
| Artificial Analysisreported | Time to first answer tokenseconds | 0.64 | 10.5 |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 26.6% | 65.8% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 10.0% | 107.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.37 | 0.29 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 26.2 | 38.5 |
Bold green marks the better value on that metric. Where a source reports ranges that overlap, the difference may not be meaningful; see the benchmark page for ranges.
Quick answers
Is Llama 3.3 70B Instruct better than gpt-oss-120b?
It depends on the task. Among models available today, on our benchmarks and licensed sources, gpt-oss-120b ranks higher for agents and tool use, reasoning and knowledge and writing.
Which is cheaper, Llama 3.3 70B Instruct or gpt-oss-120b?
gpt-oss-120b costs $0.26 per million tokens (three input to one output), against $0.64 for Llama 3.3 70B Instruct.
Which is faster, Llama 3.3 70B Instruct or gpt-oss-120b?
gpt-oss-120b writes about 200 tokens a second on its usual API, against 80 for Llama 3.3 70B Instruct (Artificial Analysis).
Run Llama 3.3 70B Instruct and gpt-oss-120b on your own prompt
Benchmarks aren't your data. Try both side by side in the playground, with the cost of every answer.