On August 21, 2026, OpenAI cut the API price of GPT-5.6 Sol, its flagship model, by more than 20%. Input tokens dropped from $5 to $4 per million. Output tokens dropped further, from $30 to $20 per million, a 33% cut. Cached input fell from $0.50 to $0.40 per million.
It's the second price cut to the GPT-5.6 family in under a month, and the first to touch Sol specifically. Terra and Luna were cut on July 30. Sol held its launch price until now.
01The New Rate
| Old rate | New rate | Change | |
|---|---|---|---|
| Input (per 1M tokens) | $5.00 | $4.00 | −20% |
| Output (per 1M tokens) | $30.00 | $20.00 | −33% |
| Cached input (per 1M tokens) | $0.50 | $0.40 | −20% |
Promotional, guaranteed at least through November 21, 2026.
At the new rate, Sol costs $4 input / $20 output, compared to Claude Opus 5's $5 input / $25 output. That makes Sol the cheaper of the two flagship models on both sides of the ledger, at least for now.
Update, Sep 24, 2026: Anthropic launched Claude Opus 5.5 on September 22, cutting its own Opus-tier rate 20% to $4 input / $20 output, the exact same headline rate as Sol's promotional price. Opus 5 remains available unchanged at $5/$25, so the comparison above still holds against that specific model, but Sol no longer undercuts every current Claude flagship option. The two now tie on sticker price; Sol's cache-read rate ($0.40, 10% of input) is still double Opus 5.5's ($0.20, 5% of input), so a cache-heavy workload favors Opus 5.5 even at an identical base rate. See the Opus 5.5 launch article for the full rate card.
Compare Sol's real cost against Claude and Gemini
See the full flagship-tier breakdown with your own token volumes02What "Promotional" Actually Means Here
OpenAI's own wording is specific: the pricing is guaranteed "at least through November 21, 2026." That's a floor, not an expiration date. The company hasn't said what happens after, whether the rate reverts to $5/$30, drops further, or gets extended. Budget for the current rate, but don't assume it holds past late November.
The cut also doesn't reach everyone. It applies to the pay-as-you-go API, Codex credits, and eligible ChatGPT Work plans. ChatGPT Pro, Plus, and Business subscriptions are unchanged, since those are seat-based rather than metered by token.
03Why Now
This is OpenAI's second Sol-adjacent pricing move in three weeks, following the July 30 cuts to Terra (20%) and Luna (80%). Taken together, all three GPT-5.6 tiers have now been discounted since the family's July 9 launch. The timing lines up with continued price competition across the industry. Anthropic, Google, and DeepSeek have each made pricing moves of their own in the same stretch.
04What This Means If You're Building on Sol
If you're already calling GPT-5.6 Sol through the API, this is a straightforward win: same model, same code, lower bill. The output cut matters more than the input cut for most workloads, since output tokens are typically the smaller share of a request but the larger share of the cost.
For anything you'd previously ruled out as too expensive for Sol-level reasoning, this is worth a second look. A task that didn't pencil out at $30 output per million may clear the bar at $20.
See what your actual usage would cost at the new rate
Run your real token volumes through the calculator