Skip to content
Model Pareto

gpt-oss-20b

OpenAIOpen weightsartificialanalysis.ai ↗
Compare with nearest rivalsCompare with…
Input
$0.07/M tok
Output
$0.18/M tok
Blended
$0.098/M tok · 3:1
Speed
181tok/s
Context
131Ktokens

Long-tail import from Artificial Analysis (2026-09-24); AA variant: gpt-oss-20b (high). Price = AA median host price. Weights ship in MXFP4 (card: "post-trained with MXFP4 quantization of the MoE weights … gpt-oss-20b … within 16GB of memory").

Rankings

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
1× B200 180GB
BF16 weights
Throughput
~11,917 tok/s
many requests batched
Price
$0.058–0.52 /M tok
busy → light use
On your own machine
RTX 5090 32GB (int4, 300+ tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • BF16 weights (21B params, 3.6B active per token) + 25% KV-cache headroom ≈ 53 GB
  • 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~11,917 output tok/s aggregate at batch 256 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~147,656 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 VerifiedEstimated
~#11 of 14 (estimated)~70.9%
  1. #1Claude Opus 596.0%
  2. #2Claude Sonnet 585.2%
  3. #10Claude Haiku 4.573.3%
  4. ~#11gpt-oss-20bEstimated70.9%
  5. #11Nemotron 3 Ultra70.7%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~#18 of 19 (estimated)~49.7%
  1. #1Claude Opus 5.589.9%
  2. #2Claude Sonnet 5.581.3%
  3. #17Muse Glimmer51.2%
  4. ~#18gpt-oss-20bEstimated49.7%
  5. #18Nex-N2.5-Mini43.8%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#28 of 31 (estimated)~45.9%
  1. #1DeepSeek V4.1 Flash74.2%
  2. #2Grok 4.772.6%
  3. #27Qwen3.8 Max (0902)51.0%
  4. ~#28gpt-oss-20bEstimated45.9%
  5. #28Qwen3.8 27B42.2%
No published score — estimated from related benchmarks
AA Coding Agent IndexEstimated
~last of 15 (estimated)below measured range (<41.1)
  1. #1Claude Opus 5.566.0
  2. #2Claude Fable 5.162.2
  3. #14GPT-6 Luna41.1
  4. ~#15gpt-oss-20bEstimated<41.1
No published score — estimated from related benchmarks

Writing & chat

MMMLUEstimated
~last of 5 (estimated)below measured range (<81.3%)
  1. #1Gemini 3.1 Pro92.6%
  2. #2Gemma 4 31B88.4%
  3. #4gpt-oss-120b81.3%
  4. ~#5gpt-oss-20bEstimated<81.3%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Verified
#48 of 531299 Elo
  1. #1Claude Opus 5.51501
  2. #2Claude Fable 5.1Few votes1496
  3. #47Molmo2-8BFew votes1305
  4. #48gpt-oss-20b1299
  5. #49Olmo 3.1 32B Instruct1294
lmarena.ai · 2026-09-13
LMArena Text (Elo)Verified
#50 of 531317 Elo
  1. #1Claude Opus 5.51509
  2. #2Claude Fable 5.11501
  3. #49Ring-flash-2.01321
  4. #50gpt-oss-20b1317
  5. #51NVIDIA Nemotron 3 Nano 30B A3B1314
lmarena.ai · 2026-09-13
EQ-Bench Creative Writing v3 (Elo)Estimated
~#24 of 26 (estimated)~1334 Elo
  1. #1GPT-6 Astra2173
  2. #2Claude Fable 5.12162
  3. #23Gemma 4 31B1368
  4. ~#24gpt-oss-20bEstimated1334
  5. #24gpt-oss-120b961
No published score — estimated from related benchmarks

Hard reasoning

Agents

OSWorld-VerifiedEstimated
~#11 of 12 (estimated)~61.7%
  1. #1Kimi K384.8%
  2. #2Qwen3.8 27B84.3%
  3. #10Muse Glimmer65.9%
  4. ~#11gpt-oss-20bEstimated61.7%
  5. #11Claude Haiku 4.550.7%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~last of 5 (estimated)below measured range (<47.5%)
  1. #1GLM-5.3 Flash59.1%
  2. #2Kimi K358.3%
  3. #4GPT-5.547.5%
  4. ~#5gpt-oss-20bEstimated<47.5%
No published score — estimated from related benchmarks
BrowseCompEstimated
~#15 of 18 (estimated)~53.5%
  1. #1Atria Dawn Preview92.5%
  2. #2GPT-6 Astra91.5%
  3. #14Inkling77.1%
  4. ~#15gpt-oss-20bEstimated53.5%
  5. #15Mistral Medium 3.548.6%
No published score — estimated from related benchmarks