GPT-6.2 Sol's pricing story is one sentence long: identical to the previous generation. But the bill is not just the rate card — the long-context surcharge, the tier multipliers and the per-task cost are three separate accounts. This article settles each one from OpenAI's model page (read directly by this site Nov 19, 2026).
| Item (per 1M tokens) | GPT-6.2 Sol | GPT-6.1 Sol | GPT-6.2 Astra |
|---|---|---|---|
| Input | $2 | $2 | $8 |
| Output | $10 | $10 | $40 |
| Cached input | $0.10 | $0.10 | $0.40 |
| Cache writes (30 minutes) | $2.50 | — | $10 |
| Cache writes (12 hours) | $5 | — | $20 |
| Long context >272K input (in / out) | $4 / $15 | — | $16 / $60 |
| Batch (in / out) | $1 / $5 | — | $4 / $20 |
| Fast mode (in / out) | $4 / $20 | — | $16 / $80 |
| Ultrafast (in / out, through Feb 28, 2027) | $6 / $30 | — | $24 / $120 |
Three points: first, $2/$10 is exactly GPT-6.1 Sol's price — the upgrade added no cost; second, Sol is a quarter of Astra on every line (input, output, cached input, cache writes, long context, Batch, Fast and Ultrafast are all a 4× relationship); third, cache writes come in two tiers — $2.50 for 30 minutes, $5 for 12 hours, against Astra's $10/$20.
The model page's rule: for requests with more than 272K input tokens, the whole request bills at surcharge rates — input and cache rates double, output costs 1.5×. For Sol that means input $2 to $4, cached input $0.10 to $0.20, output $10 to $15. CyberQ's independent Nov 18 reading ($4/$0.20/$15) matches the model-page rule exactly. Note "the whole request": it is not just the tokens above 272K that bill at the surcharge — the entire request does, so size long-document work before sending it, and lean on cache reads or split requests past the threshold.
Artificial Analysis' independent measurement (Nov 18, 2026 reading) gives cost per index task: Sol $1.01, Astra $4.00, GPT-6.1 Sol $1.27, Fable 5.2 $26.54. Sol costs about a quarter of Astra per task and about 1/26 of Fable, and about 20% less than its predecessor. AA's blended rate (a 7:2:1 cache-hit/input/output ratio) is $1.47 per 1M tokens. CyberQ's 7-question test gives a second set of equivalent costs: Sol $0.84, Astra $1.75 — the same roughly-half relationship.
Batch (50% of Standard): for batch work that isn't urgent — input $1, output $5, half the Standard rate. Note metarix quotes Batch differently (30% below Standard on a 2-hour window, 40% below on an 8-hour window) — that is a different window-pricing framing; this site follows the model page's "Batch at 50% of Standard rates". Fast (2×): input $4, output $20, for faster responses. Ultrafast (3× through Feb 28, 2027, then 6×): input $6, output $30, rising to 6× (input $12, output $60) on Mar 1, 2027. Ultrafast only pays off for the lowest latency; start everyday work on Standard. Pro 200 subscribers get Ultrafast for the first time, starting with Sol.
Included Sol usage inside ChatGPT subscriptions rose with the launch: from 15–160 to 17–180 messages per 5 hours (Plus and Standard Business seats, via metarix). Allowances vary by plan and usage intensity — the subscription page is authoritative; free users get no allowance.
Reference: OpenAI — GPT-6.2 Sol model page · Artificial Analysis — GPT-6.2 Sol model page · CyberQ — Independent agent test (rewritten and organized by ModelBrief; facts per official sources and Artificial Analysis)