# Claude June 2026 Roundup, Two Models Suspended, Two Retired, and One Billing Change Paused > Source: > Published: 2026-07-01 15:01:03+00:00 *Anthropic shipped a lot to Claude this month, more than a hundred individual updates if you count every release-note line. Most of those are small. A smaller number of them genuinely change something for a user or a developer. This is that subset, sorted by what actually matters, including the model that launched and is no longer available, the billing change that was announced and then paused, and the older models that have now been retired outright. No marketing tone, no hype, just what shipped and the honest current state.* If you’ve been trying to keep up with Anthropic’s release pace this month and feeling like the announcements blur together, this piece is for you. Anthropic shipped well over a hundred discrete updates across the Claude product family in June 2026 across the consumer apps, Claude Code, and the developer platform. The majority of those are small, internal bug fixes, performance tweaks, niche admin features. A smaller subset genuinely changes what you can do or what you pay for, and a few of them are reversals of things other guides might still tell you are happening. Here is that subset, organized by the area you care about, with the honest current state of each. Because I want this to be useful rather than promotional, I’m including everything in three buckets, what shipped, what got pulled or paused, and what is on the deprecation list with a date. The third bucket usually does not make it into roundups, and it’s the most actionable thing for anyone running Claude in production. The biggest model news of the month is also the one with the most caveats. On June 9, Anthropic launched two new top-tier models, Fable 5 and Mythos 5, as the most capable Claude models released to date. Fable 5 was the generally available version with safety classifiers built in, and Mythos 5 was the same underlying model with some safeguards lifted for restricted trusted-access use through a program called Project Glasswing. Reported gains were significant, particularly on long coding migrations, vision tasks, and finance reasoning. Here’s the honest current state, three days after that launch, on June 12, Anthropic announced it was suspending access to both Fable 5 and Mythos 5 for all customers. As of this writing, neither model is available. Anthropic has stated that all other Claude models remain unaffected. If you were planning to build on Fable 5 or Mythos 5, you can’t today, and you should plan around Opus 4.8 instead. For context on what Opus 4.8 actually is, it shipped on May 28 as the upgrade to Opus 4.7, with a 1-million-token context window as the default (it used to require a beta opt-in), roughly four times fewer code flaws overlooked compared to Opus 4.7 by Anthropic’s reported measure, more reliable tool calling, and improved multimodal reasoning across PDFs and unstructured content. Whether or not the gains are exactly as advertised in your specific workload, the 1M default context is the practical change that matters most, agent sessions that used to require summarization to keep going can now carry the full history. The deprecation list is where the actionable stuff lives, and it’s also where people get burned because they had something pinned to a model that no longer answers. On June 15, Anthropic retired Claude Sonnet 4 and Claude Opus 4. Retirement is different from deprecation, a request pinned to a retired model id fails outright with an error rather than serving a warning and continuing to work. If you are still running anything against claude-sonnet-4-20250514 or claude-opus-4-20250514, it's already broken. The migration paths Anthropic recommends are Sonnet 4.6 and Opus 4.8 respectively. Two more deprecations have dates attached. Fast mode for Claude Opus 4.7 will be removed on July 24, after which claude-opus-4-7 with speed: "fast" returns an error. Migrate to fast mode on Opus 4.8. And Claude Opus 4.1 is scheduled for retirement on August 5, with Opus 4.8 as the recommended target. Put both of those dates on your calendar if you have any code pinned to them. This one is worth a heading on its own, because several guides still describe it as live, and it isn’t. Anthropic announced a change earlier in the month that would have moved Agent SDK usage, claude -p usage, and third-party app usage onto a separate monthly credit pool starting June 15, 2026. On the day it was due to take effect, that change was paused. As of this writing, the flat-rate path for bringing your Pro or Max account to a third-party wrapper is unchanged. If you read a guide claiming the credit-pool split is live, check Anthropic's official release notes before you plan around it. Two other billing notes worth knowing about, both shipped. Since June 2, requests that end with a refusal and produce no output are no longer billed. You don’t need to opt in. If you were paying for the input tokens of requests Claude refused outright, those charges stop appearing on your bill. And the advisor tool now supports a max_tokens parameter to cap advisor output per call, useful if you have workloads that don't need a full-length advisor response. Claude Code, the command-line agent, got the largest number of updates this month by volume. Most are small fixes, but a few change the daily experience. The most useful new feature is /rewind, which lets you resume a conversation from before a /clear. If you've ever cleared context by mistake and lost important state, this is the answer. Streaming CPU use dropped by roughly 37 percent in the same update, which you notice in long sessions on a laptop. Sandbox credential blocking lets you prevent sandboxed commands from reading credential files and secret environment variables. Org-configured model restrictions show up in the model picker so administrators can lock teams to specific models. And fullscreen mode got mouse click support for the menus, with a new env variable to turn that off if you don't want it. Everything else in Claude Code this month is fixes, dozens of them across --resume, --json-schema structured output, remote MCP hangs that no longer hang, paste handling, background jobs that keep working, plugin auto-renames, OAuth retries, GitHub app setup, voice dictation on macOS, terminal rendering. These don't change what the tool can do, but they noticeably reduce the friction of using it daily. The biggest platform change is **Workload Identity Federation**, which replaces static API keys with short-lived credentials issued at request time, authenticated through the identity provider your stack already uses (AWS IAM, GCP or Kubernetes service accounts, Azure managed identity, GitHub Actions tokens, Okta, or other OIDC providers). For anyone tired of rotating API keys, this is a meaningful change. Alongside it, the platform now supports service accounts, so each workload can have its own identity, role, and audit trail rather than sharing one key. A few other platform changes worth knowing. Claude Platform on AWS launched, putting the Messages API, Files API, Batches API, Managed Agents, Agent Skills, code execution, and tool use behind native AWS endpoints with AWS billing and IAM auth. Self-hosted sandboxes for Claude Managed Agents went into public beta, letting tool execution run in an environment you control, either your own infrastructure or a provider like Cloudflare, Daytona, Modal, or Vercel, while the agent loop stays on Anthropic. Multiagent sessions and Outcomes are in public beta under the managed-agents beta header. The code execution tool now discloses a 90-second per-cell time limit in its tool description, so the model can budget long-running cells. The web search and web fetch tools added a response_inclusion parameter to drop consumed result blocks from the API response, useful for keeping agentic workflow context lean. Vault credential background refresh is now supported for OAuth credentials, and webhook event types now cover session and vault lifecycle events. A few smaller changes worth knowing about for everyday users and IT admins. Claude can now be tagged directly in Slack conversations on Team and Enterprise plans, so you can delegate a task to it in a channel and stay focused on other work. Trusted Devices for Remote Control, on Team and Enterprise plans, let admins require members to verify their device before viewing or steering local Claude Code sessions remotely. Custom roles now support admin permissions, so a member can be given access to specific admin areas like billing or privacy without making them an Owner. Enterprise-managed MCP connector access, starting with Okta integration, lets admins provision connectors once for zero-touch access on first login. On the developer side outside Anthropic, Apple’s Foundation Models framework now supports Claude on iOS 27, iPadOS 27, macOS 27, visionOS 27, and watchOS 27, so SwiftUI developers can pass typed outputs from Apple’s on-device pass into a Claude request through the framework. Claude Cowork, the desktop tool for non-developers to automate file and task management, went generally available on macOS and Windows through Claude Desktop. And Claude Design, the visual outputs product for designs, prototypes, slides, and one-pagers, launched alongside Opus 4.7 earlier this spring and continues to evolve. If you read every release note this month and tried to track what actually matters, the short list looks like this. Two top-tier models launched on June 9 and were both suspended on June 12, so plan around Opus 4.8 for now. Two older models, Sonnet 4 and Opus 4, were retired on June 15 and will error if you call them. Opus 4.1 and Opus 4.7 fast mode are on the deprecation list with dates in July and August. A billing change that would have split Agent SDK usage was announced and then paused on the day it was due to ship, so flat-rate bring-your-own-account access to wrappers is still live. Refusals that produce no output are no longer billed. Underneath all that, the platform got more enterprise-ready (Workload Identity Federation, AWS endpoints, self-hosted sandboxes, multiagent sessions, OIDC service accounts, Slack delegation, trusted devices), Claude Code got the /rewind it should have had from day one plus a long list of fixes, and a handful of capabilities crossed from beta to GA. For anyone running this in production, the immediate to-do list is short. Migrate anything still on Sonnet 4 or Opus 4 today. Put July 24 and August 5 on your calendar for the next two retirements. If you have any code pinned to Fable 5 or Mythos 5, switch back to Opus 4.8 until and unless those models become available again. And ignore any guide telling you the Agent SDK credit-pool change is live, it was paused. That’s the honest June 2026 picture, the rest is detail. *If you have been running Claude in production this month and hit something not covered here, drop a comment. The release notes tell you what shipped. The honest, on-the-ground experience of people running it is what actually matters when you are deciding what to migrate and when.* [Claude June 2026 Roundup, Two Models Suspended, Two Retired, and One Billing Change Paused](https://pub.towardsai.net/claude-june-2026-roundup-two-models-suspended-two-retired-and-one-billing-change-paused-bc3fa2fa78bb) was originally published in [Towards AI](https://pub.towardsai.net) on Medium, where people are continuing the conversation by highlighting and responding to this story.