Claude Codeupdate

What changed for Claude developers in June 2026

In brief

Six changes landed on the Claude API and in Claude Code in June 2026 that affect how you build and what you pay. Rate-limit tiers were consolidated and raised, fast mode started winding down for older Opus models, the code execution tool gained persistent state, and Claude Opus 4 and Sonnet 4 were retired. Here is each one, the date, and what to do about it.

6 min read·Claude Code

Contents

♡Sign in to save

June 2026 was a maintenance month more than a launch month. No new flagship model shipped after Fable 5 on June 9, but a string of platform and Claude Code changes landed that change rate limits, billing, and which models you can call. Here is each one with its date and the action it asks of you.

1. Rate-limit tiers consolidated and raised (June 26)

Anthropic raised rate limits across the Claude API and collapsed the old usage-tier ladder into three tiers: Start, Build, and Scale.

The change that matters most: Claude Sonnet and Claude Haiku rate limits now match Claude Opus at every tier. Previously the cheaper models often had lower ceilings, which forced some teams onto Opus just to get throughput. That distortion is gone.

Anthropic's stated outcome: most organizations move to a higher tier, no organization receives lower limits than before, and no action is required. You can see your tier and current limits in the Claude Console.

What to do: nothing required. If you were rate-limited on Sonnet or Haiku and worked around it by routing to Opus, re-check whether that workaround is still worth its cost.

2. Fast mode is winding down for older Opus models

Fast mode — the premium-priced option that returns responses roughly 2.5x faster — is being retired on the older Opus generations and consolidated onto Opus 4.8.

June 25 — Opus 4.7 fast mode deprecated. Removal is scheduled for July 24, 2026. After that date, a request to claude-opus-4-7 with speed: "fast" returns an error.
June 29 — Opus 4.6 fast mode removed. A request to claude-opus-4-6 with speed: "fast" no longer runs at fast speed or premium pricing. It runs at standard speed, is billed at standard rates, and does not return an error. The response's usage.speed field reports the speed actually used.

The June 29 behavior is the one to watch, because it fails quietly. If you have code that sets speed: "fast" on Opus 4.6 and assumes it is getting the fast path, you are now silently getting standard speed at standard price — no exception thrown.

What to do: if you depend on fast mode, migrate those calls to claude-opus-4-8, which is where fast mode now lives. See the migration guide for the model-name swap.

3. Code execution tool gained persistent state (June 18)

The code execution tool shipped a new version, code_execution_20260120, that adds REPL state persistence: variables and imports survive across cells within a session instead of resetting each time. The SDKs for Python, TypeScript, Go, Java, Ruby, PHP, and C# all support it. No beta header is required — set the tool's type to code_execution_20260120.

This version is also the minimum required for programmatic tool calling, the feature that lets Claude write a single script that calls your tools inside the sandbox instead of bouncing through the model once per call. If you are adopting programmatic tool calling, you need this version.

It is available on Claude Fable 5, Mythos 5, Opus 4.5 and newer, and Sonnet 4.5 and newer.

What to do: if you use code execution and want state to carry across cells, bump the type string. Check the model compatibility table first.

4. Claude Opus 4 and Sonnet 4 were retired (June 15)

The original Claude 4 models reached end of life:

claude-opus-4-20250514 (Opus 4)
claude-sonnet-4-20250514 (Sonnet 4)

Requests to either model ID now return an error. This was announced well in advance, but if you pinned a model ID from May 2025 and never updated it, your calls are now failing.

What to do: point Opus 4 calls at Opus 4.8 and Sonnet 4 calls at Sonnet 4.6. Researchers who need ongoing access to the retired models can request it through the External Researcher Access Program. This is the standard reason to pin a model family alias where possible and to keep a fallback path wired in.

5. Claude in Microsoft Foundry reached general availability (June 29)

Claude is now generally available in Microsoft Foundry on Azure, out of preview. Opus 4.8 and Haiku 4.5 are available through the Messages API with Azure-native identity, billing, and governance, plus optional US data zone support for eligible customers.

For teams already standardized on Azure, this means Claude access without a separate Anthropic billing relationship — the same Messages API shape you use elsewhere, routed through Foundry.

What to do: if your org runs on Azure and procurement was the blocker, this removes it. The API surface is the same; only the endpoint and auth change.

6. Claude Code app: smaller quality-of-life changes

Several changes landed in the Claude Code desktop and CLI experience through June:

Organization default models — admins can set a default model for everyone in the org, so new sessions start on the model you standardized on.
Readable session names — sessions get human-readable names at startup instead of opaque IDs, which makes the sessions list navigable.
Clickable file attachments and a smoother agents view for managing parallel work.
Trusted Devices for remote sessions (June 25) — Team and Enterprise admins can require device verification before a member connects to a remote Claude Code session.
Org-wide MCP connector provisioning — admins can provision MCP connectors for the whole organization through their identity provider, starting with Okta. Users get connector access automatically on first login instead of each person wiring it up.

None of these change your code. They change how a team is administered and how sessions are kept straight at scale.

The pattern across all six

Nothing here is a new capability you have to learn. Five of the six are lifecycle moves — tiers consolidated, fast mode migrated, old models retired, a tool version bumped, a cloud endpoint promoted to GA. The lesson they share is the same one Fable 5's suspension taught the hard way in the same month: pinned model IDs and assumed pricing paths are liabilities. The teams that spent June untouched were the ones already pointing at current model aliases with a fallback wired in.