Anthropic Weekly: Fable 5 returns, Sonnet 5 launches, and California opens the door (2026)

Anthropic's week turned on one access decision: Claude Fable 5 and Mythos 5 moved from emergency suspension back toward availability, but only with a thicker layer of cyber controls and government coordination. In the same window, Anthropic launched a cheaper frontier-ish Sonnet model, opened a science workbench, landed a statewide California procurement path, and saw the Alibaba dispute spill into workplace tool bans.

The week at a glance

Date	Development	What changed	Why it matters
Jun 30-Jul 1	Fable 5 and Mythos 5 return	Anthropic said U.S. export controls on Fable 5 and Mythos 5 were lifted on Jun 30. Fable 5 became available globally on Jul 1 across Claude Platform, Claude.ai, Claude Code, and Claude Cowork; Mythos 5 access was restored for a set of U.S. organizations after government approval. 1	The model-access crisis is no longer a simple shutdown story. It is now about who gets high-capability models, under what controls, and how quickly Anthropic can restore cloud-channel access.
Jun 30	Claude Sonnet 5 launches	Sonnet 5 became the default model for Free and Pro plans, with access across Max, Team, Enterprise, Claude Code, and the Claude Platform. API pricing starts at $2 per million input tokens and $10 per million output tokens through Aug 31, then moves to $3 and $15. 2	Anthropic is pushing agentic performance down-market while keeping Opus 4.8 and Fable 5 as higher-end options.
Jun 30	Claude Science enters beta	Anthropic introduced Claude Science, a macOS/Linux workbench for researchers with scientific artifacts, compute orchestration, specialist agents, and more than 60 curated skills and connectors. It is in beta for Pro, Max, Team, and Enterprise users. 3	This is Anthropic's clearest vertical-product move in life sciences and research workflows, not just a general chatbot feature.
Jun 29	California opens a statewide Claude path	California said state agencies may access Claude at a 50% discounted price, with free workforce training and Anthropic technical assistance. Cities and counties can use the same discounted offer. 4	Claude now has a public-sector reference deployment in Anthropic's home state, with procurement and training baked into the deal.
Jul 2	Cyber jailbreak framework published	Anthropic proposed a Cyber Jailbreak Severity scale from CJS-0 to CJS-4, mapped Fable 5 cyber requests into prohibited, high-risk dual use, low-risk dual use, and benign categories, and opened a HackerOne program for Fable 5 cyber jailbreak reports. 5	The access reversal comes with an attempt to standardize how model labs and governments talk about jailbreak risk.
Jul 3	Alibaba reportedly bans Claude Code internally	Reuters reported that Alibaba banned employees from using Claude Code at work after scrutiny of features tied to identifying China-linked users; the report said employees were being directed to Alibaba's own Qoder coding platform. 6	Last week's alleged distillation dispute is becoming an enterprise-trust problem, not only a legal or policy fight.

Access risk moved from shutdown to governed redeployment

The important change is not just that Fable 5 is back. It is that Anthropic described a path for redeployment after an unusually abrupt government intervention.

Anthropic said the Jun 12 export-control directive forced it to suspend Fable 5 and Mythos 5 globally because it could not verify user nationality in real time. After the controls were lifted on Jun 30, Fable 5 returned globally on Jul 1, while Mythos 5 remained narrower: restored to approved U.S. organizations and still tied to the Project Glasswing trusted-access program. 1

For customers, the open issue is channel restoration. Anthropic said it would re-enable Fable 5 access on AWS, Google Cloud, and Microsoft Foundry as quickly as possible, but the post did not give exact dates. That matters for enterprises that buy Claude through cloud marketplaces rather than directly through Claude.ai or Anthropic's API. 1

Anthropic also changed the safety story. It said a new classifier blocks the specific behavior described in an Amazon report in more than 99% of cases, but acknowledged the fix will increase false positives for benign coding and debugging tasks. 1 That tradeoff is now part of the product, not a footnote.

Sonnet 5 is the pricing move

Sonnet 5 is positioned as the model that brings stronger agentic behavior to cheaper, broader plans. Anthropic says it is available across all plans, is the default for Free and Pro users, and is usable through Claude Code and the Claude Platform. The launch price is meaningfully below Opus 4.8: Sonnet 5 starts at $2 per million input tokens and $10 per million output tokens through Aug 31, versus Opus 4.8 at $5 and $25 in the comparison text. 2

The safety positioning is also different from Fable. Anthropic says Sonnet 5 has lower dangerous-cyber capability than current Opus models and launches with cyber safeguards enabled by default, but with less strict safeguards than Fable 5. 2

The practical read: Sonnet 5 is the scale model. It gives Anthropic a cheaper agentic default while preserving a reason for customers to pay up for Opus or seek access to Fable when they need harder long-horizon work.

Claude Science is the vertical bet

Claude Science is more than a feature bundle. Anthropic describes it as a research environment that can analyze literature, run multi-step scientific work, generate figures and manuscripts, manage local or remote compute, and preserve auditable histories of outputs. The beta is available to Pro, Max, Team, and Enterprise users on macOS and Linux. 3

The product also has a distribution hook. Anthropic said it will support up to 50 AI-for-science projects with up to $30,000 in credits, while Modal will provide up to $2,000 in compute for select projects. Applications are open through Jul 15, with project runs scheduled from Sep 1 to Dec 1. 3

This sits directly beside the Fable/Mythos story. Anthropic is arguing that frontier scientific capability needs both access controls and specialized product packaging. Claude Science is the packaging layer.

California gives Claude a government distribution channel

California's announcement is not a splashy private-sector logo, but it may matter more operationally. The state said Claude will be the first AI productivity tool made available to all state agencies through the California Department of Technology's Statewide Information Technology Shared Services portal. The offer includes a 50% discount, free workforce training, and technical assistance from Anthropic developers; the same discount is available to cities and counties. 4

The announcement also names real usage paths: Engaged California, Poppy, cyber defense work with CDT and CalOES, DMV customer-service work, and internal workflows at the Department of Health Care Services. 4

That makes this more concrete than a memorandum-of-understanding style announcement. It gives Anthropic a government adoption case with procurement, training, and named agency workflows in one package.

Alibaba risk widened from allegation to tool ban

Reuters' Jul 3 report turns the Alibaba dispute into a user-trust issue around Claude Code. According to Reuters, Alibaba banned employees from using Claude Code at work after scrutiny of features that can help identify China-linked users. Reuters said Alibaba employees were being told to use Alibaba's own coding platform, Qoder. 6

The report ties the ban to Anthropic's earlier allegation that Alibaba illicitly extracted Claude model capabilities through distillation. Reuters also reported that neither Alibaba nor Anthropic immediately responded to requests for comment, and that Alibaba had not publicly commented on the accusation. 6

For Anthropic, this cuts both ways. The company wants stronger controls against unauthorized resale and distillation. But when those controls are visible inside developer tools, customers and competitors can frame them as surveillance or compliance risk.

What to watch next

Jul 7: Anthropic's temporary inclusion of Fable 5 for up to 50% of weekly usage limits ends for Pro, Max, Team, and select Enterprise plans; after that, usage-credit mechanics become more important. 1
Jul 15: Claude Science project applications close for the credit-supported AI-for-science program. 3
Aug 31: Sonnet 5's introductory API pricing is scheduled to end. 2
No date yet: Cloud-channel Fable 5 restoration, broader Glasswing access, and any formal response from Alibaba remain unresolved. 1 6