# Anthropic disables Claude Fable 5 and Mythos 5: what to use instead > Timeline of the shutdown of Claude Fable 5 and Mythos 5, switching models in Claude Code, and preserving working context. A historical review of June 2026. Source: https://ailoot.ai/blog/en/claude-fable-mythos-model-switch/ Canonical: https://ailoot.ai/blog/en/claude-fable-mythos-model-switch/ Language: en Publisher: Ailoot (https://ailoot.ai/) Translation: guide-008 Revision: sha256-cd4e8a0a76e9fdf267cd77f4edf08668794e8d4c964a976608ad703aec243791
Key points On June 12, 2026, the U.S. government issued an export control directive. On June 12–13, Anthropic disabled Claude Fable 5 and Mythos 5 for all customers. Among the powerful models, Claude Opus 4.8 and Sonnet 4.6 are currently available—they are unaffected. In Claude Code, one command is enough to switch models. Below: a timeline, steps in Claude Code, quotes from Anthropic, and what to do over the next 2–4 weeks.
Practical materials on vibe coding and working with models: [Ailoot documentation](https://ailoot.ai/docs/). ## What happened to Claude Fable 5 and Mythos 5 between June 9 and 13?
Key points On June 9, Anthropic released Fable 5 (a mass-market model) and Mythos 5 (for researchers). At 21:21 GMT on June 12, the U.S. government issued an export control directive. On June 12–13, Anthropic disabled both models for all customers.
A brief timeline, without the fluff: 1. ### June 9—release of Fable 5 and Mythos 5 Anthropic announces two new models. Fable 5 is aimed at a broad audience (\$10 input / \$50 output per million tokens, free on Pro, Max, Team, and Enterprise plans through June 22). Mythos 5 is limited to Project Glasswing partners and select biology researchers. According to benchmarks, Fable 5 outperforms Opus 4.8 by 10%+. 2. ### June 12 (Friday, 21:21 GMT)—U.S. government directive The U.S. administration issues an export control directive. The requirement is to prohibit access to Fable 5 and Mythos 5 for all foreign nationals, including Anthropic employees. 3. ### June 12–13—Anthropic disables access entirely Technically, it is impossible to segment accounts by citizenship (one account may be used by several people). Anthropic turns off Fable 5 and Mythos 5 for its entire customer base at once. Key point: the situation unfolded over 4 days, from release to complete shutdown. Previously, the U.S. government had never publicly halted commercial AI models—this was the first time. ## What are Fable 5 and Mythos 5—and how did they differ from Opus 4.8?
Key points Fable 5 is a public, mass-market model with mythos-class capabilities. Mythos 5 is the same model with fewer restrictions, available only to select researchers. The main difference from Opus 4.8 is the claimed 10%+ improvement on coding and agentic task benchmarks.
Fable 5 technical specifications (according to Anthropic’s official page): - Price: \$10 per million input tokens, \$50 per million output tokens. For standard, uncached Claude API usage, the official model pages checked on October 19, 2026 list Opus 4.8 at \$5/\$25: Fable is twice that price. Against the Sonnet 4.6 rate shown here (\$3/\$15), Fable is about 3.3 times as expensive. These API rates are separate from subscription promotions, caching, and batch discounts; these current prices do not establish an archived June tariff. - Included free in Pro, Max, Team, and Enterprise plans through June 22, 2026. - Available through the API, Amazon Bedrock, Google Cloud Vertex, and Microsoft Foundry. - Claimed 10%+ improvement on software engineering and knowledge work benchmarks. - Built-in safeguards—blocks responses in high-risk areas (cybersecurity, biology). Mythos 5 is the same base model, but with fewer built-in restrictions. Access: - Project Glasswing partners (the program expanded to 150+ organizations in 15+ countries as of June 2026). - Select vetted biology researchers. Key point: Mythos 5 was never a mass-market product. Disabling it for a broad audience did not affect a typical vibe coder. The impact on users is from Fable 5 specifically. ## Which Claude models work now (and which is the default)?
Key points Among the powerful models, Claude Opus 4.8 and Claude Sonnet 4.6 are available. Opus 4.8 is the flagship for complex tasks. Sonnet 4.6 is the default model in Claude Code and claude.ai on the Free and Pro plans. Sonnet 4.6 offers a million-token context window in beta for all paid subscriptions.
After disabling Fable 5, Anthropic publicly confirmed: “Access to less powerful Claude models, including the latest Claude Opus 4.8, is unaffected”. Availability snapshot: June 19, 2026. For Fable and Opus, the API-price column uses the official model-page rates as of October 19, 2026: | Model | Where it’s available | Price per 1M tokens | Context | |----|----|----|----| | Opus 4.8 | API, Bedrock, Vertex, Claude Code, claude.ai (Max, Team, Enterprise) | \$5 / \$25 | up to 1M in beta | | Sonnet 4.6 | API, Claude Code, claude.ai (Free, Pro, Max, Team, Enterprise—default) | \$3 / \$15 | up to 1M in beta for all paid plans | | Haiku 4.5 | API, Claude Code, claude.ai | \$0.80 / \$4 | 200K | | Fable 5 | DISABLED | \- | \- | | Mythos 5 | DISABLED | \- | \- | For a typical vibe coder, this means nothing critical has broken. Sonnet 4.6 has been the default model in Claude Code since late February 2026. In Anthropic’s benchmarks, it is described as “capable of delivering results that previously required an Opus-class model”—in other words, it is sufficient for most vibe-coding tasks. ## How to switch back in Claude Code: 2 minutes
Key points In Claude Code, enter `/model` at the command line to see the list of available models. Choose `claude-opus-4-8` for complex tasks, or leave Sonnet 4.6 as the default. If Fable 5 was listed in `.claude/settings.json`, remove it manually.
The full process in 4 steps: 1. ### Open \`/model\` in Claude Code In your Claude Code work session, type `/model` to open the list of available models. Fable 5 will no longer appear there after June 13. Choose `claude-opus-4-8` (for demanding architecture work) or `claude-sonnet-4-6` (the default, for fast coding). 2. ### Check \`.claude/settings.json\` If you previously explicitly set Fable 5 in the `model` field, open the file and replace its value with `claude-opus-4-8` or `claude-sonnet-4-6`. Without this edit, Claude Code will fail at session startup with the error "model not available". 3. ### Check system prompts and subagents If Fable 5 is explicitly specified in `.claude/agents/*.md` or `CLAUDE.md` (through `model:` frontmatter in a subagent file), replace it with an available model. A subagent with a nonexistent model silently drops out of planning. 4. ### Restart Claude Code `/exit` and start it again. Check that the session starts and `/model` shows a working model at the top. If it starts, the tool is ready to use. On macOS, also check that `claude` is on your PATH—after reinstalls, the link sometimes gets lost: ```bash which claude claude --version ``` If `which claude` returns nothing, download a fresh version by following the instructions in Anthropic’s docs. An older version may not know about models added after the CLI release. ## What Anthropic says—three direct quotes
Key points Anthropic considers the directive a mistake and says it is working to restore access. The company disagrees that a narrow vulnerability should lead to the recall of a commercial model. It also says the government should have leverage to block deployments, but through a transparent process based on facts.
Anthropic issued public comments within the first day after the directive. In its own words:

We apologize to our customers for the disruption. We believe this is a misunderstanding, and we are working to restore access as quickly as possible.

— Anthropic

We disagree that the discovery of a narrow potential vulnerability should become the reason for recalling a commercial model that has been deployed to hundreds of millions of people.

— Anthropic

We believe the government should be able to block unsafe deployments—through a legislative process that is transparent, fair, understandable, and grounded in technical facts.

— Anthropic
Anthropic’s position is constructive. The company does not dispute the right to regulate AI. But it considers this particular directive a mistake. Practical materials: [Ailoot documentation](https://ailoot.ai/docs/). ## What changes when you choose a model What changes in the regional scenario described as of June 19, 2026: - If you had a Pro/Max subscription and used Fable 5 during the first 2–3 days after its release, Claude Code will now simply switch to Sonnet 4.6 (the default) or Opus 4.8 (if you choose it through `/model`). There will be no noticeable quality drop on 90% of tasks. - If you were planning to switch to Fable 5 after June 22 (when the free period ends and you have to pay \$10/\$50), the question has resolved itself. You are going back to Sonnet 4.6 with a million-token context window in beta. - If you were exploring the use of Mythos 5, forget it: access is only through Project Glasswing, subject to vetting. ## A vibe coder’s strategy for the next 2–4 weeks
Key points Anthropic is clearly working to restore access. But the forecast is unrealistic—switch Claude Code to Opus 4.8 or Sonnet 4.6, keep working, and don’t rewrite your product based on speculation. And checkpoint your CLAUDE.md so any future model change won’t disrupt your context.
Three priorities for the coming weeks: 1. ### Switch models and forget about Fable 5 Don’t wait for it to return. Anthropic says it is “working to restore access,” but there is no specific date. Follow the instructions in the Claude Code section above and continue working on Opus 4.8 / Sonnet 4.6. They cover 95% of needs. 2. ### Checkpoint your CLAUDE.md A good CLAUDE.md is insurance against model changes. If the context in `business/` and `ai-clone/` is assembled properly, Claude of any version will produce more or less the same results. If your instructions depend on “Fable-class capabilities,” rewrite them in more general terms so they work with Opus 4.8 and any hypothetical future model. For a detailed analysis, see the guide [“Context Engineering 2026”](https://ailoot.ai/blog/en/context-engineering/). 3. ### Don’t move to Codex or Cursor just because of this story Anthropic has had one shutdown this quarter, and it is temporary. Other providers have their own policies and risks. It is worth moving if the problem is ongoing. You don’t change tools because of one incident. For an analysis of alternatives, see [“Claude Code vs Codex”](https://ailoot.ai/blog/en/codex-vs-claude-code-choice/) and [“Cursor vs Claude Code”](https://ailoot.ai/blog/en/cursor-vs-claude-code/). 4. ### Follow Anthropic’s news—but not every hour Anthropic will release an update on access. You can adapt when it comes out. Until then, hourly monitoring wastes time and disrupts your work. Checking @AnthropicAI on X or anthropic.com/news once every 2–3 days is enough. An example from the practice described: a project was built in 3–4 hours on Opus 4 in February 2026. Two months later, the instructions were partially rewritten in CLAUDE.md for Opus 4.5, and then again for 4.7. Each model change takes 15–30 minutes of edits. No rebuild is needed. The same would apply to Fable 5: if it had been used in production, you could have rolled back to Opus 4.8 in 20 minutes without losing functionality. ## What to expect in the coming months
Key points Anthropic will keep pressing for dialogue with the regulator, most likely through a narrow adjustment to Fable 5’s safeguards and a rerelease. At the same time, the company may release new models that are not covered by the directive. Forecast: Fable 5 returns within 2–6 weeks in an updated version.
Possible developments: - Anthropic updates Fable 5’s safeguards (closes the bypass technique discovered by the regulator) and releases Fable 5.1 or 5.5. This is the most likely scenario because it directly addresses the directive’s formal requirement. - Anthropic releases a new model outside the mythos class, but close in quality and without objections from the regulator. It will be named something else to avoid legally falling under the directive. Unlikely; it would take months. - Anthropic challenges the directive in court—but that is a process lasting a year or more, and it does not help current customers. What NOT to expect: Fable 5 returning in its previous form. If you expect it back within a week—no, regulators do not reverse course that quickly.
### Sources - [Anthropic. News & announcements](https://www.anthropic.com/news) - official releases and statements. - [Anthropic on X](https://x.com/AnthropicAI) - public response to the directive. - [AWS Blog. Anthropic Claude Fable 5 on AWS](https://aws.amazon.com/blogs/aws/anthropic-claude-fable-5-on-aws-mythos-class-capabilities-with-built-in-safeguards-now-available/) - Fable 5 technical specifications. - [Claude API Docs. Introducing Claude Fable 5 and Claude Mythos 5](https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5-and-claude-mythos-5) - official model documentation. - [Guide: Context Engineering 2026](https://ailoot.ai/blog/en/context-engineering/) - protection against model changes through CLAUDE.md. - [Guide: Codex vs Claude Code](https://ailoot.ai/blog/en/codex-vs-claude-code-choice/) - alternative tools.
This story is a reason to put together a working vibe-coding setup so you depend less on a specific model. More materials and tools: [Ailoot](https://ailoot.ai/). ## Questions about Anthropic disables Claude Fable 5 and Mythos 5: what to use instead | Ailoot ### Which settings should I check after switching an unavailable model? Check `/model`, the `model` field in `.claude/settings.json`, and explicit model declarations in `.claude/agents/*.md`. Restart Claude Code and confirm that the intended available model is selected. ### Are API token prices the same as a subscription allowance? No. The API comparison uses standard input/output prices per million tokens; the article’s temporary subscription promotion was a separate condition. Verify the rate and access terms for the provider and date you will actually use. ## Ailoot · The company and its project portfolio Ailoot combines established operating experience, a large client base, stable service and ongoing KPI control. The company and its project portfolio: millions of customers; monthly sales in the billions. Ailoot brings that operating experience to API access for AI features in customer products, agents, applications and automations.