Search was never the hard part.
Doing something with the results was.
On 17 February 2026, Claude's web search tool and programmatic tool calling reached general availability, with code execution, web fetch, tool search, tool examples and the memory tool all clearing beta the same day. The quieter detail is the one worth training on: code execution is now free when it runs alongside a search.
Ask Claude to look something up and it searches. Ask it to look something up and then calculate, compare or reshape what it found, and until 17 February 2026 that second half often meant an extra step, and sometimes an extra charge. General availability and a free pairing with code execution changed both.
Seven releases cleared beta on the same day
Alongside the launch of Claude Sonnet 4.6 on 17 February 2026, Anthropic moved the web search tool and programmatic tool calling to general availability, dropping the beta headers they previously required. Code execution, web fetch, tool search, tool examples and the memory tool reached the same milestone that day. None of these tools were new individually. What changed was reliability, the thing that actually decides whether a team builds a real workflow on a feature or leaves it as a demo.
The pairing that is actually free
The detail worth training people on is smaller than the general availability headline: code execution became free of charge specifically when used together with web search or web fetch. That removes the cost calculation teams used to run before letting Claude write and run a script against something it just found online.
What the combination unlocks
Claude searching for current information, then writing and running code to calculate, compare or reshape what it found, in a single pass, with no person manually bridging the gap between finding the data and doing something useful with it. Research summaries that need real figures cross-checked, competitive pricing comparisons, and dataset cleaning that starts with finding the latest version of a file all benefit directly.
Why general availability changes the maths
Beta status came with caveats most teams quietly built around: extra error handling, a mental note not to rely on it for anything client-facing, a habit of treating the output as provisional. General availability removes that hedge, and the free pairing with web search removes the cost hedge too, which is exactly the combination that turns a feature people tolerate into one they actually plan workflows around.
The habit most teams have not built yet
The gap we see most often is not technical. People still ask Claude to search, read the results themselves, then separately ask it to process something, purely out of habit from before the tools worked well together. Asking for the combined workflow directly, search for this, then calculate or reshape it this way, in one request, is a small change in phrasing that saves a genuinely meaningful amount of back-and-forth.
Teach people to ask for the search-then-process workflow in one request rather than two manual steps. It is a small phrasing change that suits watching someone do it live far better than reading about it in documentation.
Things people
usually ask.
The web search tool and programmatic tool calling, with their beta headers removed, alongside code execution, web fetch, tool search, tool examples and the memory tool, all on the same day Claude Sonnet 4.6 launched.
Specifically when it is used together with web search or web fetch in the same request. That pairing is what dropped the cost barrier, not code execution on its own in every context.
Claude searching for current information, then writing and running code to calculate, compare or reshape what it found, in one pass, without a person manually bridging the gap between finding the data and doing something with it.