Simon Willison: May was a good month for Anthropic (and expensive for AI)

Simon Willison published on June 1st the May edition of his monthly newsletter for sponsors. The summary he previews on his blog is brief but nuanced: AI has become more expensive, Anthropic had one of its best recent months, and May's model launches left many feeling that expectations were not met.

Willison is one of the most influential and candid voices in the developer tools ecosystem. His newsletter is not marketing or a press release summary: it is the diary of an engineer who uses these technologies daily and evaluates them without compromise. That is why it is worth paying attention even to the headlines he shares publicly.

AI gets more expensive: a signal worth watching

That AI costs rose in May is not a minor data point. For months, the dominant trend went the opposite direction: cheaper models, larger context windows, prices per token falling. If Willison highlights the cost increase as one of the month's headlines, it probably is not an anecdotal price adjustment from a minor provider.

This type of cost pressure affects teams unevenly. Teams that have built heavy pipelines on APIs, with chained calls, long contexts, or intensive agent use, notice it first. For those working with Claude Opus 4.7 and its 1M token windows in production, each price increase has a real impact on project viability.

Anthropic had a good month, despite the models

The combination Willison describes is striking: Anthropic did well in May, but the model releases were "somewhat disappointing." This suggests Anthropic's strong month did not necessarily rest on technical surprises in the form of new versions, but probably on positioning, enterprise adoption, deals, or product advances that cannot be measured by benchmarks alone.

It is a distinction worth keeping in mind. The Claude ecosystem has been maturing for months in layers that go beyond the base model: Claude Code with its support for skills, sub-agents, hooks, and plugins; the consolidation of MCP as a real integration standard with external tools; growing adoption in enterprise environments. A "good month" might mean those layers are delivering results, even if the model itself did not make a dramatic leap.

Datasette Agent: a project to watch

Among the launches of his own that Willison mentions, Datasette Agent stands out, a tool that combines his well-known data exploration project with agentic capabilities. Willison describes making "significant progress" on Datasette during May, and the agent appears to be the most visible piece of that work.

For those working with tabular data and looking to integrate natural language query flows, this type of open-source project built on Claude APIs has immediate practical value. It is not a large company product with SLA and support: it is an engineer publishing what he builds and explaining how he does it, making it a reliable technical reference.

The newsletter as an ecosystem barometer

The full May edition is reserved for GitHub sponsors at 10 dollars a month, with immediate access to the private archive. The April edition is already publicly available as a sample, giving an idea of the level of detail Willison offers each month.

For those of us following the Claude ecosystem from a technical and editorial perspective, this newsletter is one of the few resources where someone with sound judgment makes a real monthly assessment: what worked, what disappointed, what he is using and why. In a space full of announcements and noise, that has value.

---

Editorial Note: That a methodical developer like Willison flags AI cost increases as a May headline deserves follow-up. If the trend confirms in June, it will have practical consequences for many projects that currently take certain cost margins for granted.