GPT-5.6 Sol
Last verified August 23, 2026 · OpenAI pricing ↗$4.00/1M input · $20.00/1M output · 1.1M context · OpenAI
Count Tokens
—
Tokens
—
Input Cost
—
Output Cost
Estimate Monthly Cost
Monthly Cost Estimator
Quick:
<$0.0001/mo
Pick a preset above or enter custom usage
Alternatives to GPT-5.6 Sol
Pricing Details
OpenAIPRICE CUT 2026-08-21: OpenAI dropped Sol from $5.00/$0.50/$30.00 to $4.00 input, $0.40 cached, $20.00 output, announced on its developer forum as 'over 20%' and 'for the next 3 months'. Input and every input-derived line fell exactly 20% (x0.800); output and every output-derived line fell exactly 33.3% (x0.6667). No multiplier changed: cache reads stay 90% off, cache writes 1.25x input ($5.00), Batch and Flex half ($2.00/$10.00), Fast a flat 2x ($8.00/$40.00), and the >272K rule still 2x input / 1.5x output ($8.00/$30.00; Batch/Flex long $4.00/$15.00). Fast long context works out at $16.00/$60.00 by the two published multipliers but is not printed as its own cell. PROMOTIONAL BUT NOT DATED, AND THERE IS NO priceSchedule ON THIS ENTRY ON PURPOSE: the model page says the rate is 'available at least through November 21, 2026'. That is a FLOOR, not an expiry - the earliest date OpenAI has reserved the right to move the price, not a date on which it will. OpenAI has published no rate for November 22 and no end date at all. This entry previously carried priceSchedule effectiveUntil 2026-11-21 reverting to $5.00/$30.00, which was an inference and not a published fact; it would have raised the calculator's Sol estimate 40% on 2026-11-22 for a change OpenAI never announced, so it was removed on 2026-08-28 for the same reason Sonnet 5's cancelled schedule was removed. Do not re-add one without a published OpenAI end date to cite. Covers the pay-as-you-go API plus eligible ChatGPT Work and Codex credits; Pro, Plus and Business subscription usage explicitly excluded. Terra, Luna and Cyber did NOT move on 2026-08-21. Against Claude Opus 5 ($5.00/$0.50/$25.00, batch $2.50/$12.50) the new card is exactly 0.800000 on input, cached input, output and both batch legs, so Sol is 20% cheaper than Opus 5 on any token mix; pre-cut it matched on input and was 1.2x on output, and if the promo lapses it returns to 1.03x-1.19x depending on output share. CHANNELS as of 2026-08-23 (one model, eight prices, 2.75x on input): OpenRouter and Vercel $2.00/$10.00 (a further 50% off the new list, no published end date); OpenAI direct and Bedrock Global CRIS $4.00/$0.40/$20.00; Bedrock In-Region and Geo $4.40/$0.44/$22.00 (the 1.1000x multiplier survived the cut exactly); Requesty $4.50/$27.00 after 10% off a stale $5/$30 list; Azure Global Standard still $5.00/$30.00 and Data Zone $5.50/$33.00, with every Sol meter in the Azure Retail Prices API still dated 2026-07-01, i.e. Azure did not act and is now 1.25x input / 1.50x output of OpenAI; Bedrock Daybreak Blue still $5.50/$33.00, so the cyber-program approval penalty grew from 1.10x to 1.375x input and 1.65x output. In-family spreads after the cut: Terra 2.0x input / 1.667x output (was 2.5x/2.5x, so Terra went from 40% to 56% of Sol on a 3:1 blend), Luna 20x/16.667x (was 25x), Cyber 3.125x/3.75x (was 2.5x). Family output spread Luna $1.20 to Cyber $75.00 = 62.5x. Prior context, still accurate: only tier with the max and ultra reasoning modes; ultra runs parallel subagents at the same per-token rate but spends more tokens per task. Fast mode was a RENAME of Priority Processing on 2026-07-30, service_tier accepts both 'priority' and 'fast', with published SLAs of 99.9% uptime and 99% of requests above 80 output tok/s (Terra Fast 70, Luna Fast 100). The >272K surcharge reprices the WHOLE request, not the excess, and since 2026-08-05 long-context prompts can also run in Fast mode, so the two rules combine. Note the >272K standard input rate ($8.00) still equals the Fast short-context input rate while delivering none of the speed. Terminal-Bench 2.1: 88.8% base, 91.9% ultra. Sol on Cerebras at up to 750 tok/s from July. Artificial Analysis (read 2026-08-23, Index v4.1.1, 9 evals): Intelligence Index 60.93, 16,879 output tokens per task, $1.2312 per task. AA has NOT applied the 08-21 cut - its published cost figures still run on $5.00/$0.50/$30.00, confirmed by dividing its published Sol answer cost by its published answer tokens and landing on exactly $30.00/M. Recomputed at the new card that is about $0.9175 a task. Opus 5 scores 63.05 at $2.3369 a task on 40,249 output tokens, i.e. it spends 2.38x Sol's output tokens per task, so per finished task Opus 5 costs 1.90x Sol at the old card and 2.55x at the new one, against a 0.8x rate-card ratio. Index points per dollar: Opus 5 27.0, Sol 49.5 old and about 66.4 new. Its Terminal-Bench 3.0 run, published at $3,950.90, reprices to about $3,068.73 and from $30.86 to about $23.97 per solved task; the drop is bounded 20.00%-23.80% since cache reads were 71.5% of that bill.
Input / 1M tokens
$4
Output / 1M tokens
$20
Context Window
1.1M
Max Output
128K
Price History
| Date | Input /1M | Output /1M | Change |
|---|---|---|---|
| 2026-08-21 | $5.00 → $4.00 | $30.00 → $20.00 | ↓ -20% |
Frequently Asked Questions
Common questions about GPT-5.6 Sol pricing and usage
Read More About GPT-5.6 Sol
Three labs shipped a security-gated model in three days. Anthropic published a full rate card for one its own cyber programme cannot sell you yet, Google published no number at all, and OpenAI published a pointer.
September 5, 2026 · 12 min read
GPT-6 Astra's rate card is Claude Fable 5's rate card, to the cent, on every line both companies publish. Anthropic moved one of those lines two days before Astra shipped, and it is now the only line where the two flagships disagree.
September 4, 2026 · 13 min read
Eight AI API discounts expire on a published date between September 7 and January 1. Half of them are video models nobody is tracking, the loudest one is the smallest, and a ninth that everybody lists has no end date at all.
August 28, 2026 · 14 min read