About
gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for...
Related Models
Pricing
Input
$0.03
per 1M tokens
Output
$0.14
per 1M tokens
Blended
$0.06
per 1M tokens
Cheaper than 89% of models. Median price is $0.54/1M tokens.
Cost Calculator
Tokens per day1M
100K100M
Daily
$0.06
Monthly
$1.70
vs. Similar Models
Nemotron 3 Nano Omni 30B A3B ReasoningQ:0.0
$0.13+131%
Mistral: Mistral Medium 3.1Q:-0.1
$0.80+1310%
Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)Q:+0.2
$0.17+208%
GPT-5 (ChatGPT)Q:+0.4
$3.44+5958%
Performance
238
tokens/sec
Faster than 93% of models
0.59
seconds
Faster than 78% of models
8.99
seconds
Faster than 42% of models
Market Median
95 tok/s
152% faster
Median TTFT
1.11s
46% faster
Throughput/Dollar
4196
tok/s per $/1M
Speed Comparison
Sarvam 30B (high)
241 tok/s+1%
Qwen3.5 Omni Flash
243 tok/s+2%
LFM2.5-8B-A1B
232 tok/s-3%
Context Window
131K
tokens
Larger than 27% of models
Max Output
131K
tokens
100% of context
Benchmarks
MMLU-Pro
74.8%
GPQA Diamond
68.8%
HLE
9.8%
LiveCodeBench
77.7%
SciCode
34.4%
TerminalBench Hard
10.6%
MATH-500Not evaluated
AIMENot evaluated
AIME 2025
89.3%
IFBench
65.1%
Long Context Recall
30.7%
Tau2
60.2%
Market AverageTop Score
Open Source
apache-2.020BGGUF / GPTQ / AWQ
Downloads
7.0M
Likes
4.7K
VRAM (FP16)
24-48 GB
GPU
A6000 / M3 Ultra
