GPT-5.6 Sol API Pricing: OpenAI Cuts Rates to $4/$20 (Aug 21 – Nov 21, 2026)

Published August 22, 2026By ABD Legacy LLC
GPT-5.6 Sol API pricing OpenAI pricing AI model cost

What happened. On Aug 21, 2026, OpenAI cut API and credit pricing for its flagship top-tier model, GPT-5.6 Sol, by over 20% for the next three months — the first price cut on Sol since launch. New rates: $4 per 1M input tokens and $20 per 1M output tokens, with cached input at $0.40 per 1M (down from $5/$30 and $0.50).

The window

The promotional pricing is available Aug 21 – Nov 21, 2026 (at least through November 21). It applies to the API and eligible ChatGPT Work and Codex credits; ChatGPT Pro, Plus, and Business subscription usage is unchanged.

What it means for agency cost math

Input-heavy agent workloads drop roughly 20%+ at the new rate — a 10K-in/2K-out Sol task falls from ~$0.110 to ~$0.080, and a 100K-in/5K-out run from ~$0.650 to ~$0.500. Sol remains the premium frontier tier, so routing guidance is unchanged; the cut narrows the premium gap to open-weight and mid-tier models.

Re-run your per-task estimates with the verified $4/$20 rates in the AI Agency Pricing Calculator, and see Find AI Agency's model routing recommendations for when premium models are worth the price.

The Jul 27 – Aug 14 discount experiment: why this cut matters

Before the Sol promo, OpenAI ran a controlled discount experiment on its other two GPT-5.6 models — Terra and Luna — from July 27 to August 14, 2026 via OpenRouter (a 50% discount layered on list prices). Usage followed price hard: daily Luna token usage jumped 13.8x and Terra rose 5.6x during the window, while Sol — kept at list price as the control — rose only 1.1x. Nearly a third of users who tried a discounted model kept using it after the discounts expired, and Terra/Luna went from 0.7% to 7.8% of OpenRouter tokens. OpenRouter frames the data as pricing-elasticity evidence of Jevons paradox: lower token prices drove disproportionately higher usage.[1][2]

Those temporary cuts became permanent list prices on Jul 30: Terra dropped 20% to $2/$12 (from $2.50/$15) and Luna dropped 80% to $0.20/$1.20 (from $1/$6) per 1M tokens. Sol was the only GPT-5.6 model left at list price — which is why its own $4/$20 promo (announced Aug 21) is the last piece of the family to move. For the full breakdown of what the experiment means for agency pricing, see OpenAI's discount experiment: what 13.8x usage growth means for agency pricing.

See the new GPT-5.6 Sol rate in your cost model

Open the AI Agency Pricing Calculator →

Sol is modeled at the verified promo rate: $4 input / $20 output / $0.40 cached input per 1M tokens.

Frequently asked questions

What is GPT-5.6 Sol API pricing after the cut?

GPT-5.6 Sol now costs $4 per 1M input tokens and $20 per 1M output tokens, with cached input at $0.40 per 1M — down from $5/$30 and $0.50. The promotional rate is valid Aug 21 through at least Nov 21, 2026, per OpenAI's Aug 21, 2026 announcement.

How long does the GPT-5.6 Sol promotional pricing last?

The GPT-5.6 Sol promotional pricing is available at least through November 21, 2026 (the three-month window from the Aug 21, 2026 announcement). After the window, pricing is expected — but not guaranteed — to revert to $5/$30.

Does the GPT-5.6 Sol price cut apply to ChatGPT subscriptions?

No. The cut applies to the API and to eligible ChatGPT Work and Codex credits. ChatGPT Pro, Plus, and Business subscription usage remains unchanged.

Related reading

Sources

Accuracy note: Rates verified 2026-08-28 against the official OpenAI community announcement, OpenAI developer docs, AWS Bedrock announcement, and the OpenRouter discount-window analysis (research brief t_5c8841ca: 7 sources, 18 evidence quotes). Cost-per-task deltas are derived arithmetic from the published rates (~20% input / ~33% output cuts). The promo window is "at least through Nov 21, 2026"; post-window pricing is expected but not guaranteed to revert to $5/$30. Terra/Luna list prices ($2/$12, $0.20/$1.20) are the Jul 30, 2026 official rates; the usage multiples (Luna 13.8x, Terra 5.6x, Sol 1.1x control) and ~1/3 retention are OpenRouter's measured discount-window data, attributed as such. Prompts over 272K input tokens are billed 2x input / 1.5x output; cache writes bill at 1.25x the uncached input rate.