WWhichAITry via Vercel AI Gateway
Models/OpenAI
OpenAI · released 2025-09open

GPT-OSS 20B

OpenAI's open-weights return.

context
128K
output max
8K
input /M
$0.15
output /M
$0.60
modalities
T
Verdict

What it's actually good at

Strengths

  • +Single-GPU inference
  • +Open weights
  • +Cheap on hosts

Weaknesses

  • Below flagship quality
  • No vision

Our scores

reasoning
7.0
coding
7.0
writing
7.0
speed
9.0
value
10.0
OpenAI's open-weights release. Runs on a single GPU, surprisingly capable, and finally free to self-host.
Pricing

Cost at common volumes

Monthly volumeInputOutputEst. monthly
Side project1M tok0.3M tok$0.33
Growing app50M tok15M tok$16.50
Production500M tok150M tok$165.00
Scale5000M tok1500M tok$1650.00

Estimates assume uncached input. Prompt caching and batch APIs can cut this by 50–90% for many workloads. Use the calculator →

Head-to-head

Compare GPT-OSS 20B