GPT-OSS 20B vs Llama 4 405B
OpenAI's open-weights return. Meanwhile, llama 4 405b: open-weights heavyweight. self-host or run anywhere.
Side-by-side
Specs
| Dimension | GPT-OSS 20B | Llama 4 405B |
|---|---|---|
| Provider | OpenAI | Meta |
| Released | 2025-09 | 2025-07 |
| Context window | 128K | ✓256K |
| Output max | 8K | ✓16K |
| Input /M | ✓$0.15 | $2.70 |
| Output /M | ✓$0.60 | $2.70 |
| Modalities | text | ✓text, image |
| Open weights | yes | yes |
Scorecard
Dimension-by-dimension
GPT-OSS 20B
Llama 4 405B
reasoning
7.0
8.0
coding
7.0
8.0
writing
7.0
8.0
speed
9.0
6.0
value
10.0
8.0
Verdict
Which wins, by use case
| Use case | Winner | Why |
|---|---|---|
| coding | Llama 4 405B | a coin flip on our weighted coding score |
| writing | Llama 4 405B | a coin flip on our weighted writing score |
| chat | GPT-OSS 20B | a decisive lead on our weighted chat score |
| agents | Llama 4 405B | a coin flip on our weighted agents score |
| summarization | GPT-OSS 20B | a decisive lead on our weighted summarization score |
| translation | GPT-OSS 20B | a clear edge on our weighted translation score |
| reasoning | Llama 4 405B | a clear edge on our weighted reasoning score |
| research | GPT-OSS 20B | a coin flip on our weighted research score |
| vision and multimodal | GPT-OSS 20B | a coin flip on our weighted vision score |
| Cheapest AI models for bulk workloads | GPT-OSS 20B | a decisive lead on our weighted cheap-bulk score |
Bottom line
Our take
Pick GPT-OSS 20B if you need: single-gpu inference, open weights.
Pick Llama 4 405B if you need: open weights — self-host, no vendor lock-in.
At a 500M-input / 150M-output monthly volume, GPT-OSS 20B costs roughly $165 vs Llama 4 405B at $1755. Use our calculator to plug in your own numbers.
Keep browsing