Fable 5 becomes standard on Claude Max, Copilot code review gets personalized, HeyGen launches AI Clipping

July 18, 2026 is marked by the permanent availability of Claude Fable 5 on the Max and Team Premium plans, a new security release of Claude Code CLI, and a batch of developer tooling updates: the FrontierCode leaderboard at Cognition, expanded customization for Copilot code review, and the arrival of AI Clipping in the HeyGen API.

Fable 5 becomes standard access on the Max and Team Premium plans

July 18 — Anthropic announces that, starting July 20, access to Claude Fable 5 will become a standard feature of the Max and Team Premium plans, up to 50% of the limits of those subscriptions. Users on Pro and Team Standard plans will keep access through paid usage credits and will receive a one-time $100 credit as compensation.

Plan	Access to Fable 5 starting July 20
Max, Team Premium	Standard, up to 50% of the plan limits
Pro, Team Standard	Paid usage credits + one-time $100 credit

This is a change in nature, not just in date: for several weeks, Anthropic had been extending access to Fable 5 step by step — the last known deadline ran until July 19. This time, access is no longer being temporarily renewed but permanently built into the offering.

“Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits. Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit. Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.” — @claudeai on X

Anthropic acknowledges, in the rest of the same thread, that Fable availability has been “frustrating” in recent weeks, and presents this move to standard access as a way to clarify what each subscription includes.

Claude Code keeps its weekly limits raised by 50% until August 19

July 18 — Alongside the announcement about Fable 5, the ClaudeDevs account says that Claude Code weekly limits remain 50% above their usual level, this time until August 19, for all Pro, Max, Team, and seat-based Enterprise users. The benefit applies automatically, with no action required from users: no plan change or subscription switch is needed to take advantage of it, including for organizations managing multiple Enterprise seats.

🔗 @ClaudeDevs on X

Claude Code CLI v2.1.214: permission security and EndConversation tool

July 18 — Version v2.1.214 brings a series of security fixes to the permissions system: a bypass fixed on Windows PowerShell 5.1 sessions, secure failure (fail closed) for Bash checks on file descriptor redirection forms, and systematic confirmation now required for commands longer than 10,000 characters (previously run automatically). It also adds the EndConversation tool, which allows Claude to end a session on its own in the event of a highly abusive user or a jailbreak attempt — a capability already present on claude.ai since 2025. Several other fixes concern background sessions: a daemon accidentally deleting its successor’s socket, and ended sessions that could not be deleted via claude rm.

🔗 Claude Code Changelog v2.1.214

Cognition launches the FrontierCode leaderboard

July 17 — Cognition, the maker of the Devin coding agent, publishes a dedicated page for its FrontierCode leaderboard: “a page that tracks which models write code you’d actually merge.” The ranking does not measure the raw ability of models to generate code, but the likelihood that the code will actually be merged as-is into a real project — an angle closer to enterprise usage than classic benchmarks. The page centralizes the scores of all tested models, including competitors such as Grok 4.5 and Inkling, with full methodology and task examples. Cognition had published a first iteration of the benchmark in early July (FrontierCode 1.1); this page marks the move to a permanent and public tracking format.

🔗 @cognition on X

Copilot code review: expanded customization and configurability

July 17 — GitHub expands the customization of Copilot code review. Custom instructions (copilot-instructions.md, *.instructions.md, agent skills, AGENTS.md) are now read from the pull request’s head branch rather than the base branch, which makes it possible to test them directly in a feature branch without merging them first. Support also extends to REVIEW.md, GEMINI.md and CLAUDE.md files. A dedicated copilot-code-review.yml file lets you configure the execution environment independently of Copilot cloud agent, and the review now runs by default behind a firewall restricting network access — not yet supported on self-hosted runners. Organization-level runner configuration, which used to be shared between the two features, is now split into two separate sections.

🔗 Copilot code review — customization and configurability

HeyGen launches AI Clipping in its API

July 18 — HeyGen adds AI Clipping to its API: a single call turns a long recording into short clips ready to publish on social networks. The system automatically evaluates the segments of a source video to extract the passages deemed most relevant, with a score assigned to each generated clip. The feature is part of HeyGen’s suite of agentic video production tools, alongside the Batch APIs (July 17) and the HyperFrames project, for which the company has detailed a new capability every day since early July. Technical documentation is available at developers.heygen.com/ai-clipping.

🔗 AI Clipping — @HeyGen on X

Briefs

ClaudeDevs — feedback on 6 months of Claude Platform APIs — three Claude Platform leaders (Katelyn Lesse, Angela Jiang, Jess Yan) look back at the APIs added over the last six months to build and deploy agents, and at the patterns observed in production. 🔗 @ClaudeDevs on X
Cursor improves its Slack integration — the agent now shares a plan before starting a task, supports multi-repository environments, and can read and write across multiple channels and threads. 🔗 Cursor Changelog
Copilot metrics per repository reach general availability — two new endpoints in the Copilot usage metrics API provide a daily per-repository breakdown of Copilot coding agent and code review activity. 🔗 GitHub Changelog
GitHub Copilot app integrated into the usage metrics API — daily and 28-day reports now include usage of the GitHub Copilot app (active users, sessions, tokens). 🔗 GitHub Changelog
GPT-5.5 promotional pricing for Copilot Max and Pro+ — 60% off for Copilot Max users and 40% for Pro+ users, from July 18 00:00 UTC to July 20 00:00 UTC, with no code or opt-in required. 🔗 @github on X
Pi officially integrates xAI subscriptions — the open source Pi agent (harness) now supports SuperGrok and X Premium subscriptions, with access to Grok 4.5. 🔗 @pidotdev on X
Kimi Business Membership for businesses — Moonshot AI launches an enterprise subscription (minimum 5 seats, annual billing, bank transfer, dedicated support) built on the benefits of the Kimi Allegretto plan. 🔗 @Kimi_Moonshot on X
Codex CLI 0.144.6 — fix for GPT-5.6 context windows — hotfix that corrects the announced context window (272,000 tokens) for GPT-5.6 Sol, Terra, and Luna in the CLI’s built-in instructions. 🔗 Codex CLI v0.144.6

What this means

Demand management is becoming a product issue in its own right at Anthropic. The move from weekly extensions of Fable 5 access to standard inclusion in the Max and Team Premium plans, combined with the continued expansion of Claude Code limits through August 19, shows that Anthropic is now trying to stabilize how its offerings are understood rather than managing capacity shortages on a case-by-case basis. It is also a vote of confidence in available compute capacity, after several weeks in which the company itself acknowledged that demand was hard to anticipate.

Agentic coding tooling is being organized around reliability and verifiability. The security fixes in Claude Code CLI v2.1.214 for the permissions system, Cognition’s new FrontierCode leaderboard, which measures code that is actually mergeable rather than raw scores, and the expanded customization of Copilot code review (reading instructions from the PR branch, firewall by default) all point to the same concern: making code agents safer to run and their results easier to verify before production integration.

GitHub is refining how Copilot usage is measured as adoption grows. Repository-level metrics granularity and the integration of the Copilot app into the usage metrics API let organizations identify precisely where AI is generating value, a prerequisite for targeting adoption efforts rather than rolling them out blindly.

Video production keeps becoming automated at the end of the pipeline. With AI Clipping, HeyGen is tackling the step that follows filming or generation: turning long-form content into short formats ready for social networks, in a single API call — a logical extension of its strategy of agentic production tools rolled out in daily increments since early July.