Meta · Workhorse · Open weights
Llama 4 Scout
A multimodal open-weight model with a 10M-token context window, by far the largest available, using 17B active parameters. The catch is hardware: roughly 55GB of VRAM to run it.
Strong at
- -10M token context
- -Open weights
- -Multimodal
Typical use
- -Very long document sets
- -Self-hosted deployments
Watch out for
- -Needs roughly 55GB VRAM
Same tier
What else to look at
| Model | Developer | Context | In / 1M | Out / 1M |
|---|---|---|---|---|
| Claude Sonnet 5 | Anthropic | 1M | $3 | $15 |
| Claude Sonnet 4.6 | Anthropic | 1M | $3 | $15 |
| GPT-5.6 Terra | OpenAI | 1M | $2 | $12 |
| GPT-5.4 | OpenAI | - | $2.50 | $15 |
| Gemini 3.6 Flash | 1M | $1.50 | $7.50 |
Verified 2026-08-07.