Reference
Model context windows & prices
The two numbers that decide whether a model fits your job: how much it can read at once, and what it charges for reading and writing. Prices are indicative Standard-tier list prices per 1M tokens, checked 2026-11-04 — providers reprice often, so verify before budgeting a product on them.
| Model | Context window (tokens) | Input / 1M | Output / 1M |
|---|---|---|---|
| GPT-6 Astra | 1,050,000 | $10 | $50 |
| GPT-6.1 Sol | 1,050,000 | $2 | $10 |
| GPT-6.1 Terra | 1,050,000 | $1 | $5 |
| GPT-6 Luna | 1,050,000 | $0.10 | $0.50 |
| Claude Opus 5.5 | 1,000,000 | $4 | $20 |
| Claude Sonnet 5.5 | 1,000,000 | $2 | $10 |
| Claude Haiku 5.5 | 1,000,000 | $0.10 | $0.50 |
| Gemini 3.8 Flash | 1,048,576 | $0.75 | $3.75 |
Tiered and dated rates: Claude Haiku 5.5 costs $0.10/$0.50 only for prompts up to 100,000 tokens ($0.50/$2.50 above that). Gemini 3.8 Flash costs $0.75/$3.75 through December 31, 2026, then $1.50/$7.50 from January 1, 2027.
How to read this table
A 1,050,000-token window (the GPT-6 family) holds roughly 787,500 English words — about ten novels. But “fits” is not “works well”: accuracy on the middle of a very long context is worse than on a short one, so the practical window for careful work is smaller than the advertised one. Cost is the other half: output tokens run 5× input prices on every current model here, so a model that writes long replies can cost more than a “cheaper” model with a shorter window. And the spread is enormous: GPT-6 Luna charges $0.10 per million input tokens, GPT-6 Astra $10 — a 100× gap for the same token count.
Try your own text in the token counter to see how much of each window it uses.