OpenAI Cuts GPT-5.6 Sol Price — Impact & What It Means

OpenAI has cut the direct API price of GPT-5.6 Sol, its frontier model for complex professional work. Input tokens are 20% cheaper and output tokens are 33.3% cheaper for standard short-context requests. OpenAI describes this as promotional pricing and says it will remain available at least through November 21, 2026. It has not published the rate that will apply afterward.

The dated old-versus-new table above and the live pricing modules are maintained separately: the first preserves this change event, while the chart and calculator use today’s canonical rates. See the full model card on our OpenAI pricing page or enter your own traffic in the AI token calculator.

What changed

The reduction applies to direct GPT-5.6 Sol API token pricing. Cached-input reads remain one tenth of uncached input, while cache writes remain 1.25 times the uncached input rate. Long-context requests above 272,000 input tokens still trigger higher rates for the full request, but those rates have fallen with the promotion as well.

The generated old-versus-new table above carries the exact Standard short-context and long-context rates. Keeping those values in the dated event data prevents this article’s prose from becoming a second, stale price card when the promotion ends.

The price change is also rolling out across eligible ChatGPT Work and Codex credit plans. OpenAI says Pro, Plus, and Business subscription pricing and included usage remain unchanged. This is therefore an API-and-credits promotion, not a broad ChatGPT subscription price cut.

OpenAI had already reduced Terra by 20% and Luna by 80% in late July. The new Sol cut completes a cheaper GPT-5.6 ladder without changing the models’ documented capabilities, context windows, or endpoints.

What this means for API bills

Output-heavy coding agents, research systems, and long-form generation gain the most because output received the larger percentage cut. A workload with the same token mix should now see a lower Sol token subtotal immediately; tools, web-search calls, regional processing, taxes, and platform fees remain separate.

A representative monthly workload with 100M uncached input tokens and 20M output tokens saves 27.3%. An output-heavy workload with equal input and output volumes saves 31.4%. Use the calculator above for the current bill rather than carrying these percentage examples into a fixed budget.

The discount also narrows Sol’s price gap with Claude Opus 5, while Claude Fable 5 remains the higher-priced frontier route in the generated comparison. Price alone does not establish which model is cheaper per successful task. Compare accepted-result rate, retries, tool calls, latency, and human correction on the same workload.

The promotional window matters for annual planning. Treat the lower bill as temporary until OpenAI publishes the later rate. Keep both promotional and pre-promotion scenarios in forecasts so a reversion does not break product margins.

OpenRouter’s separate 50% promotion now sits on top of the lower OpenAI base rate. Its eligible OpenAI Standard route therefore costs half the direct card shown by the live modules. That is a route price, not the direct OpenAI card; provider, service tier, routing, and fallback behavior still belong in every forecast.

Who benefits—and who does not

Existing Sol users benefit immediately without a model migration. Teams that reserved Sol for hard coding, planning, cybersecurity, or agentic steps can process more of the same workload per budget. Developers using eligible Codex or ChatGPT Work credits should watch their account rollout rather than assume the change is already visible everywhere.

Pro, Plus, and Business subscribers do not receive a lower monthly subscription price or a larger published usage allowance from this announcement. Teams already routing routine work to Luna or Terra should not move it to Sol solely because Sol is cheaper; the lower tiers can still win when they pass the same evaluation.

What developers should do now

1. Recalculate August-through-November forecasts with the live rate, then retain a post-promotion rollback scenario.
2. Check invoices or usage dashboards rather than inferring billing from the model name, especially for credits and third-party routes.
3. Keep stable instructions and tool schemas in cacheable prefixes to preserve the cached-input discount.
4. Route defined, high-volume steps to Luna or Terra and reserve Sol for requests where its quality improves cost per accepted task.
5. Re-run the same production evaluation before changing traffic shares; the price changed, not the model’s documented behavior.

For broader alternatives, compare the Anthropic pricing page and DeepSeek pricing page. Our earlier OpenRouter Sol discount analysis remains route-specific; direct OpenAI pricing has now changed separately.

Labs inclusion

AI Pricing Guru Labs already includes GPT-5.6 Sol. The fresh August 22 run attempted all 49 deterministic text tasks, returned no endpoint errors, and scored 49/49. For its 3,426 prompt tokens and 901 completion tokens, the discounted OpenRouter route billed exactly half the canonical direct Standard estimate. The public result keeps the full timestamped cost record.

That compact machine-graded result does not establish quality for long context, safety, creative work, production agents, or every coding workload. The price changed, but the model-quality claim still needs a workload-matched evaluation.

Bottom line

OpenAI’s Sol promotion is a meaningful developer price cut: input is 20% lower and output is 33.3% lower through at least November 21. It improves the economics of output-heavy frontier workloads without requiring a migration, but it should not be treated as a permanent annual rate yet. Update forecasts now, verify credit-plan rollout, and keep routing by cost per accepted result.

Sources: OpenAI’s official GPT-5.6 Sol model documentation, developer pricing page, the OpenAI developer announcement, Reuters, and the live AI Pricing Guru API dataset. Rates and promotion terms reverified August 24, 2026 after the pricing update resurfaced on Hacker News.