The frontier pricing floor was supposed to hold. Sol launched in July at $5/M input and $30/M output, and OpenAI held that rate while cutting everything below it. Luna dropped 80% on July 30. Terra dropped 20%. Sol stayed put. The message was clear enough: mid-tier gets commoditized, the flagship holds value.

Then last Friday, the flagship moved. GPT-5.6 Sol dropped to $4/M input and $20/M output — 20% on input, 33% on output — with cached input falling from $0.50 to $0.40/M. OpenAI called it a promotional rate, valid through at least November 21, 2026. The cuts apply to the API and to credit pools for ChatGPT Work and Codex. Consumer subscriptions — Plus, Pro, Business — are unchanged.

Reuters reported that Sol now undercuts Claude Opus 5 on both input and output tokens. That's new. For the first time since Anthropic launched its frontier lineup, the most expensive OpenAI model is cheaper than the most expensive Anthropic model.

The word "promotional" is doing a lot of work in this announcement.

Source spread

• OpenAI — Advancing the price-performance frontier with GPT-5.6 — hype. Official announcement. Emphasizes developer accessibility, price-performance framing. No statement on what happens after November 21.
• Reuters via Investing.com — builder. Breaking coverage with the competitive context: Sol now cheaper than Anthropic's flagship on both token types.
• BenchLM — OpenAI API pricing August 2026 — builder. Third-party pricing tracker with the full current table across all GPT-5.6 tiers and context cache.

Pros & cons

What's real:

• The output cut matters more than the input cut. Most long-context and agentic workloads generate substantially more output than input. A 33% output cut on Sol changes the math on more production workloads than the 20% input cut does.
• Sol is now, for the first time, priced below Anthropic's flagship model. That's a real competitive signal. Builders who benchmarked Sol against Opus 5 and chose Opus 5 on price should rerun the comparison.
• The cuts cover ChatGPT Work and Codex credits, not just raw API. If you're running agentic workflows through either of those, your effective cost dropped as of Friday without any migration.

What deserves a side-eye:

• "Promotional" is not language companies use when the new price is the real price. It's language companies use when they want the option to raise prices back without it counting as a reversal. The word is in there for a reason.
• November 21 is approximately three months out. That's also approximately when OpenAI's next major model — Astra, now paused for safety review — might be positioned for launch. A new model is a clean justification for a price reset.
• This is the third pricing move in under a month. Luna −80%, Terra −20%, now Sol −20% to −33%. The pace of the cuts suggests either competitive panic or a genuine structural shift in inference cost. Those two things look identical from the outside.

Further reading

• OpenAI — Advancing the price-performance frontier with GPT-5.6 — official pricing announcement with full table
• OpenAI — Previewing GPT-5.6 Sol — original Sol launch for model capability context
• Reuters via Investing.com — Reuters breaking coverage with competitive context
• BenchLM — OpenAI API pricing August 2026 — third-party pricing tracker for side-by-side comparison