SurfingBearHome › AI News › Tool & Product Updates
OpenAI cuts developer pricing for its top GPT-5.6 Sol model by ‘more than 20%’ — but only for three months
Announced 21 August. API pricing comes down, but Pro, Plus and Business subscriptions are unchanged. And the fact that the cut is explicitly time-boxed to three months is the most important condition for anyone evaluating it.
What changed
According to Reuters, OpenAI said on Friday 21 August that it would cut developer pricing for its frontier model GPT-5.6 Sol by more than 20% for the next three months. Reuters attributed the move to intensifying competition from Anthropic and from Chinese AI models.
- On OpenAI’s price list, GPT-5.6 Sol is $4 per million input tokens and $20 per million output tokens on standard short-context usage. The previous prices were $5 and $30 respectively.
- The cut applies to the API, and the company said it would also roll through to eligible plans carrying credits for ChatGPT Work, its agentic product, and Codex, its coding tool.
One thing to watch is how the size of the cut is expressed. The company’s stated figure is ‘more than 20%’, while the price list shows input going $5 → $4 and output $30 → $20. Input and output fell by different proportions, so your actual saving depends on the input-to-output token ratio of your own workload.
What did not change — subscription pricing
The company was explicit that Pro, Plus and Business subscription pricing is unchanged. In other words, this cut has nothing to do with the per-seat cost of your staff using ChatGPT. It applies only to what you spend wiring the API into products and workflows.
Part of a run that started last month with the smaller models
Reuters noted this is not the first such cut. OpenAI had already cut prices on its lower-tier models late last month — mid-tier GPT-5.6 Terra by 20% and low-cost Luna by 80%. This announcement brings the same treatment to the top of the range.
Competitor pricing, as printed by Reuters
Every figure in the table below is as stated in the Reuters article. Nothing on this page has been calculated or converted. GPT-5.6 Sol pricing is on a standard short-context basis.
| Model | Input (1M tokens) | Output (1M tokens) |
|---|---|---|
| GPT-5.6 Sol (after cut) | $4 | $20 |
| GPT-5.6 Sol (before cut) | $5 | $30 |
| Claude Fable 5 (Anthropic list price) | $10 | $50 |
| Claude Opus 5 (Anthropic list price) | $5 | $25 |
In the same week, x.ai announced Grok 4.6 on the Google Enterprise Agent Platform at $2 input and $6 output per million tokens. We covered that announcement in the weekly briefing.
What this means for Korean teams — SurfingBear editorial
The first principle is not to treat this cut as your budget baseline. The company itself put ‘three months’ on it. What the price is after that is not in the announcement, and we do not know either. Building next year’s budget or your product cost model around $4/$20 means building on a number with nothing behind it. We would suggest writing the pre-cut pricing ($5/$30) into your plan and recording the discount separately as temporary headroom.
Second, what this genuinely changes is the range of what you can afford to try. Until now the common pattern has been to use the top model for demos and then drop to a mid-tier model for the actual rollout. A three-month window suits the opposite experiment — take one workload already running on a mid-tier model, move it up to the frontier model, and find out in numbers whether the quality difference re-justifies the cost. Three months is long enough to finish that experiment and reach a conclusion.
Third, a price list is not an invoice. The table is standard short-context pricing. Real cost turns on prompt length, retry counts, cache hit rates, and how many times your agent calls the model within a single request. A 20% lower unit price still produces a bigger bill if call volume climbs. The work that actually matters right now is not rebuilding your vendor comparison sheet — it is measuring your own workload’s monthly token usage.
Fourth, do not misread the frozen subscription pricing. Your internal ChatGPT seat costs are unaffected by this announcement. The two belong in separate budget lines.
To work out how much unit-price movement actually affects the economics of a rollout, try the AI ROI calculator. To scope which workload to wire up to an API first, try the automation finder. If you are at the stage of pulling several models into one stack with contracting and billing handled, see AI product sourcing.
Sources
- Reuters, “OpenAI cuts developer pricing for frontier GPT-5.6 Sol model by more than 20%”, 2026.08.21 — read the original
- x.ai, “Grok 4.6 on Google Enterprise Agent Platform”, 2026.08.21 — read the original
The size of the cut (‘more than 20%’), its duration (three months), the before-and-after pricing ($5/$30 → $4/$20), the subscription freeze, the application to ChatGPT Work and Codex credits, last month’s Terra 20% and Luna 80% cuts, and the Claude Fable 5 and Opus 5 list prices in the table are all quoted as printed in source 1 (Reuters). Grok 4.6 pricing is as stated in source 2 (x.ai announcement). No figure on this page — discount percentage, saving, or otherwise — has been calculated by us, and no currency conversion has been applied. The ‘What this means for Korean teams’ section is SurfingBear editorial interpretation, not a claim made in the original reporting.
To reach a conclusion inside three months
We help you pick one workload, wire it to the API, and measure monthly token usage alongside quality. Managed by a Korean PM.