OpenAI Reduces Price of GPT-5.6 Sol by 33%, Now Cheaper than Claude Opus 5

OpenAI announced on Thursday, August 21, a price reduction for the GPT-5.6 Sol model in its commercial API, lowering the launch cost by 33% and positioning the Sol below Anthropic's Claude Opus 5 for the first time since both models were launched. The adjustment, described by the company as promotional, remains in effect until November 21, 2026.

The new rates set the entry price at $4 per million tokens, down from the previous $5, and the exit price at $20 per million, down from $30. The Claude Opus 5, launched by Anthropic on July 24, 2026, costs $5 per million tokens on entry and $25 on exit. With the new prices, the Sol is now cheaper on both entry and exit compared to its main direct competitor in the advanced reasoning models segment.

This cut is not isolated. In July, OpenAI had already reduced the price of the Terra model by 20% and the Luna model by 80%, both positioned in lower-capacity tiers of the GPT-5.6 family. The pattern suggests systemic pressure on margins across the hierarchy of models, rather than a one-time initiative for the Sol. Sam Altman, CEO of OpenAI, did not publicly comment on the August move, but stated in July that the company intended to continue lowering prices as inference efficiency improved.

The reduction specifically applies to access via API, the Codex environment, and the ChatGPT Work plan aimed at businesses. The consumer plans ChatGPT Pro, Plus, and Business are unaffected. This distinction is relevant for CIOs pricing internal development budgets: a team processing 500 million tokens of output per month would save $5,000 monthly with the new prices, equivalent to approximately $60,000 annually before any prompt optimization.

In Japan, where large integrators like Fujitsu and NTT Data maintain scale API contracts for back-office automation, the cost differential with equivalent models from Anthropic and Chinese Moonshot AI is expected to reignite contract negotiations in Q4 2026. In the UK, the Financial Conduct Authority is already monitoring inference costs as a relevant variable for the sustainability of automated decision systems in financial services, making API price reductions a regulatory factor, not just a commercial one.

In Brazil, where the adoption of foundational model APIs in fintechs and financial service companies grew throughout 2025, the adjustment comes at a time of budget revision for the second half of the year. Companies that tested the Sol in pilot projects during the first half now have a more favorable argument for scaling in production, especially for complex reasoning tasks that require top-tier models, rather than lower-capacity solutions.

From a strategic perspective, the move reveals tension in the premium pricing premise. The Sol was positioned by OpenAI as a top-tier reasoning model, capable of processing structured tasks with more reasoning steps, justifying the higher initial price compared to Anthropic's Opus 5. With the price inversion, OpenAI signals a preference for volume and market share over maintaining the price differential as a perceived quality argument.

Anthropic has not publicly commented on the competitive move. DeepSeek, which maintains open-source reasoning models with significantly lower inference costs than the Sol even after the cut, remains the most economical alternative for companies willing to host models locally or via cloud providers offering DeepSeek inference as a managed service.

For CIOs and CTOs making model vendor decisions in the second half of 2026, the Sol's price cut alters the total cost of ownership calculation but does not resolve issues of latency, regional API availability, and compliance with data sovereignty requirements. These factors remain the primary differentiators in corporate bids across Europe and Asia, where regulators require that inference data does not cross certain geographic boundaries.