gpt-oss-120b
- Input
- $0.15/M tok
- Output
- $0.50/M tok
- Blended
- $0.237/M tok · 3:1
- Speed
- 245tok/s
- Context
- 131Ktokens
- Cost per task
- ~$0.21/taskest.
Open weights; price is the median third-party host price (Artificial Analysis). Weights ship in MXFP4 (card: "post-trained with MXFP4 quantization of the MoE weights, making gpt-oss-120b run on a single 80GB GPU").
Rankings
Coding
Mixed#102 / 137rank by quality33.0±5.3quality / 1004/7 benchmarks measuredWriting & chat
Mixed#92 / 162rank by quality36.8±3.4quality / 1005/5 benchmarks measuredHard reasoning
Verified#105 / 218rank by quality45.7±3.3quality / 1004/4 benchmarks measuredAgents
Verified#104 / 187rank by quality27.8±4.3quality / 1004/7 benchmarks measured
Self-hostingEstimated
How we estimate →What it would cost to run these open weights yourself on rented GPUs, compared with the API price above.
- Hardware
- 2× H200 141GB
- FP8 weights
- Throughput
- ~10,270 tok/s
- many requests batched
- Price
- $0.076–0.62 /M tok
- busy → light use
- On your own machine
- Mac Studio M5 Max 128GB (int4, ~170 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- FP8 weights (117B params, 5.1B active per token) + 25% KV-cache headroom ≈ 146 GB
- 2× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
- ~10,270 output tok/s aggregate at batch 369 (bandwidth-bound); MoE compute scales with active params
- Blended 3:1 input:output; prefill ~183,349 tok/s
- Low = 75% utilization at the low GPU price; high = 20% at the high price
Benchmark scores
~ italic, dashed = no published score yet, estimated from related benchmarks.
Coding
SWE-bench VerifiedLab-reported
#18 of 2062.4%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- #17Intern-S2-Preview (35B-A3B)64.0%
- #18gpt-oss-120b62.4%
- #19NVIDIA Nemotron-Labs Dory1.0 4B59.5%
arxiv.org · 2025-08-05
SWE-bench Pro (Public)Estimated
~#23 of 24 (estimated)~45.3%
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #22Muse Glimmer51.2%
- ~#23gpt-oss-120bEstimated45.3%
- #23Nex-N2.5-Mini43.8%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#42 of 46 (estimated)~37.1%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #41Laguna S 2.140.4%
- ~#42gpt-oss-120bEstimated37.1%
- #42Nex-N2.5-Mini36.1%
No published score — estimated from related benchmarks
LiveCodeBenchVerified
#8 of 2787.8%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #7Nemotron 3.5 Super VL88.2%
- #8gpt-oss-120b87.8%
- #9Nemotron 3.5 Super VL NVFP487.4%
artificialanalysis.ai · 2026-09-24
SciCodeVerified
#98 of 11834.0%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #97NVIDIA Nemotron-Labs Dory1.0 4B34.9%
- #98gpt-oss-120b34.0%
- #99Devstral 232.8%
artificialanalysis.ai · 2026-09-24
AA Coding Agent IndexEstimated
~#25 of 26 (estimated)~28.4
- #1Claude Opus 5.561.7
- #2Claude Opus 559.7
- #24GPT-6 Luna31.1
- ~#25gpt-oss-120bEstimated28.4
- #25Muse Spark Open26.3
No published score — estimated from related benchmarks
Writing & chat
IFBenchVerified
#40 of 13969.0%
- #1MiniMax-M382.9%
- #2Qwen3.8 2.4T A95B82.8%
- #39Apriel-v1.6-15B-Thinker69.1%
- #40gpt-oss-120b69.0%
- #41Mistral Medium 3.568.8%
artificialanalysis.ai · 2026-09-24
LMArena Text, Non-English (Elo)Verified
#46 of 561338 Elo
- #1Gemini 4 ArgonFew votes1512
- #2Claude Opus 5.51497
- #45Trinity Large Thinking1350
- #46gpt-oss-120b1338
- #47INTELLECT-3Few votes1327
lmarena.ai · 2026-09-24
LMArena Text (Elo)Verified
#47 of 561352 Elo
- #1Gemini 4 Argon1525
- #2Claude Opus 5.51507
- #46INTELLECT-31356
- #47gpt-oss-120b1352
- #48Llama 3.1 Nemotron Ultra 253B v1Few votes1348
lmarena.ai · 2026-09-24
#24 of 25961 Elo
eqbench.com · 2026-09-24
Hard reasoning
GPQA DiamondVerified
#83 of 19278.2%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #82Nova 2.0 Pro Preview78.5%
- #83gpt-oss-120b78.2%
- #84LongCat 2.078.0%
artificialanalysis.ai · 2026-09-24
Humanity's Last ExamVerified
#82 of 19819.6%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #81GPT-5.5 Instant (June 2026)19.9%
- #82gpt-oss-120b19.6%
- #83Gemma 4 26B A4B19.3%
artificialanalysis.ai · 2026-09-24
Agents
τ²-bench (Telecom)Verified
#44 of 11965.8%
- #1Step 3.7 Flash98.5%
- #2DeepSeek-V4-Pro-0813-nvfp4-DSpark98.3%
- #43Nemotron 3 Super 120B A12B67.8%
- #44gpt-oss-120b65.8%
- #45Qwen3.5 0.8B65.2%
artificialanalysis.ai · 2026-09-24
OSWorld-VerifiedEstimated
~#12 of 15 (estimated)~61.4%
- #1Kimi K384.8%
- #2Qwen3.8 27B84.3%
- #11Muse Glimmer65.9%
- ~#12gpt-oss-120bEstimated61.4%
- #12Nemotron 3.5 Super VL NVFP452.3%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~last of 5 (estimated)below measured range (<47.5%)
No published score — estimated from related benchmarks
BrowseCompEstimated
~#19 of 22 (estimated)~58.8%
- #1Atria Dawn Preview92.5%
- #2GPT-6 Astra91.5%
- #18Nemotron 3.5 Bolt 74B-A9B70.0%
- ~#19gpt-oss-120bEstimated58.8%
- #19Mistral Medium 3.548.6%
No published score — estimated from related benchmarks
AutomationBench-AAVerified
#79 of 840.2%
- #1Claude Fable 5.276.4%
- #2Muse Wonder Full71.3%
- #77Granite 4.2 3B0.4%
- #79gpt-oss-120b0.2%
- #79gpt-oss-20b0.2%
artificialanalysis.ai · 2026-09-24