# OpenAI Cuts GPT-5.6 Sol Prices by More Than 20 Percent, Undercutting Claude Opus 5

Input falls from $5 to $4 and output from $30 to $20 per million tokens for three months, the second repricing of the 5.6 family inside a month.

- Published: 2026-08-23T06:19:43.709Z
- Canonical: https://polylog.news/ai/2026-08-23/openai-cuts-gpt-5-6-sol-prices-by-more-than-20-percent-under
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [OpenAI Developer Community](https://community.openai.com/t/20-price-reduction-for-gpt-5-6-sol-api-codex-credits-and-chatgpt-work/1391726), [Storyboard18](https://www.storyboard18.com/amp/brand-marketing/openai-slashes-gpt-5-6-sol-api-prices-by-more-than-20-amid-competition-108453.htm)

OpenAI cut the price of its top reasoning model on Friday. GPT-5.6 Sol now costs [$4 per million input tokens and $20 per million output tokens](https://community.openai.com/t/20-price-reduction-for-gpt-5-6-sol-api-codex-credits-and-chatgpt-work/1391726), down from $5 and $30. The reduction applies to direct application programming interface (API) calls and to eligible Codex and ChatGPT Work credits. Consumer subscription prices are unchanged. OpenAI described the cut as a promotion running for three months, through November 21.

The output-token cut is the more aggressive move. Reducing output pricing from $30 to $20 per million tokens is a 33 percent cut on the rate that dominates total billing for reasoning models, because these models generate long hidden reasoning steps before producing an answer. That places Sol below [Claude Opus 5, which lists at $5 and $25 per million tokens](https://www.edenai.co/post/claude-opus-5-benchmarks-scores-pricing-routing), on both input and output. A [Russian-language technical channel tracking the change](https://t.me/ai_machinelearning_big_data/10745) noted it is the second move on the 5.6 family within a month.

OpenAI attributes the cut to efficiency gains rather than competitive pressure, which is the standard framing and cannot be checked from outside. What can be observed is the timing. Anthropic reported a $65 billion revenue run rate last week and is preparing a public listing, open-weight coding models are now scoring within a few points of the closed frontier on independent harnesses, and [Meta opened its own paid model API in July](https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/) at $1.25 and $4.25 per million tokens. Three months is also a conveniently short commitment. A promotional window lets OpenAI buy usage share now and restore list prices later without announcing an increase.

For teams building on the API, the practical note is that a temporary price is not a planning input. Anyone rebuilding unit economics around $4 and $20 should assume the old rates return on November 22 unless OpenAI says otherwise.

## What this means

Frontier vendors are now competing on the output-token rate specifically, because agentic and reasoning workloads generate far more output than input, and that is where switching decisions get made. OpenAI gains developer volume and lock-in during the window. Anthropic is exposed on price at the exact moment it is marketing itself to public-market investors on revenue growth, and any response either compresses its margin or concedes the price-sensitive tier. Inference providers with thin gross margins lose either way, since they must match a promotion set by a vendor that owns its own serving stack.

## What to watch

- Whether Anthropic responds with a price change on Opus 5 or holds list price and competes on capability, which will show how much pricing power it believes it has ahead of a listing.
- What happens on November 22, since quietly extending the promotion would indicate the cut was a permanent repricing described as temporary.
- Whether cached-input and batch rates move as well, because those set the real cost floor for high-volume production workloads rather than the headline rate.