OpenAI cuts GPT-6 Sol and Luna API prices by 50 percent

By Splitfeed EditorialLast updated Sep 26, 2026, 20:18 UTC

On September 22 and 23, 2026, OpenAI announced the release of two new models, GPT-6 Sol and GPT-6 Luna, with API token prices cut by 50 percent across both. A third model in the family, GPT-6 Astra, was made available for production agent workloads inside Microsoft Foundry, the enterprise deployment platform. On the same platform, Dun & Bradstreet commercial data was integrated into Microsoft AI agents, extending the release beyond a simple pricing announcement into a broader enterprise stack update. The New Stack noted that prompt caching may represent a larger practical cost reduction than the headline price cut, depending on how developers structure their requests.

The 50 percent price reduction continues a trend of rapidly falling frontier model costs, which has consequences for how quickly businesses can justify deploying AI at scale. The benchmark caveat is the key tension here: the lower-priced Sol and Luna models appear to involve performance trade-offs compared to higher tiers, meaning the effective cost saving depends on whether a given use case can tolerate those trade-offs. The Astra model, positioned for production agents in Microsoft Foundry, sits above that tier, suggesting OpenAI and Microsoft are trying to segment the market between cost-sensitive developers and enterprise customers who need higher capability. The Dun & Bradstreet integration adds a data layer that makes the Microsoft Foundry release specifically relevant to commercial and financial enterprise workflows.

The Pro-AI camp has not yet published reactions to this story. Based on typical positions, that camp would likely emphasize the democratizing effect of lower API prices and point to enterprise integrations as evidence of maturing, practical deployment. The Anti-AI camp, as represented by himay on Mastodon, is not engaging with this specific pricing news at all; instead, the account linked to a Mother Jones piece about legal case documents and claims that OpenAI may have committed what one executive described as the largest theft of labor in human history, framing the broader OpenAI story around copyright and worker harm rather than product pricing. From the Middle Ground, Paul Guinnessy on Bluesky acknowledged the Microsoft integration but noted that the features feel buried inside Microsoft apps, and expressed frustration that the company does not provide tutorials alongside its webinars, a comment pointing to a gap between announced capability and accessible implementation.

The argument most worth tracking is whether the benchmark performance gap between Sol and Luna and higher-tier models proves meaningful for real enterprise workloads, and how the ongoing legal questions around OpenAI's training practices affect enterprise adoption decisions. Any court filings or rulings in the referenced copyright case would shift the conversation around the company's releases considerably.