OpenAI halves GPT-6 prices, Claude Opus 5.5 arrives 20% cheaper, three AI providers go down on one morning, Ireland fines Google €403 million, and EU AI Act enf
Five stories from the past few weeks that matter to anyone buying, running or regulating AI, with our take on each.
On September 23, OpenAI cut GPT-6 Sol to $2 per million input tokens and $10 per million output tokens, down from $4 and $20, and GPT-6 Luna to $0.10 and $0.50, down from $0.20 and $1.20. OpenAI credited better caching and inference, and said the new rates have no expiration date. Source.
Our take: A lower list price is welcome, but it does not lower the cost to finish the work on its own. The gap between the cheapest and the strongest models keeps widening, which makes choosing the right model for each step worth more, not less. Rework still costs the same whatever the token price. That is the case for CompletionPrism™.
On September 22, Anthropic released Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens, about 20% below Claude Opus 5. Source.
Our take: Two price moves in two days is the new normal. A company that has to benchmark, approve and retrain people every time a model changes will always be a quarter behind. New models should be priced and used the day they win, without anyone changing how they work.
On September 3, Claude had a partial outage across Claude.ai, Claude Code and the API lasting just over three hours. ChatGPT and Codex were unavailable for some users for about 34 minutes after what OpenAI called a routing error, and Grok went down after an outage at an xAI compute center. Source.
Our take: Any company running on one AI provider stopped work that morning, and any software vendor whose features depend on one provider let its customers down. Model-level failover belongs in the plan, and in every SaaS vendor's sales story. See the Embedded License.
On September 21, Ireland's Data Protection Commission fined Google Ireland €403 million over how it processed location data, finding failures of lawful basis, transparency and retention, and ordered Google to bring the processing into compliance within six months. Source.
Our take: Lawful basis, transparency and retention are the same three questions every marketing database faces. Contacts gathered without a clear basis, or kept with no limit, carry the same risk at a smaller scale. That is why RelationLens shares a contact only when the employee approves it, and forgets what nobody reviews.
Since August 2, the European Commission, through its AI Office, holds formal investigative and enforcement powers over providers of general-purpose AI models, including the power to request information and technical documentation. Source.
Our take: When a regulator asks what an AI system decided, and why, a log someone could have edited is a weak answer. Companies using AI in hiring, credit and healthcare decisions need records that prove themselves. See InferTrust™.
In the News comes out every week. Subscribe from the Insights page to get it by email.