ai-powered-markdown-translatorArticle translated from fr to en with gpt-5.4-mini.
On July 12, the news is largely centered on AI-assisted development tools: Anthropic extends Fable 5 access by one week and raises Claude Code limits by 50%, while Amp (Sourcegraph) says it has switched its default agent from Opus to GPT-5.6, cutting average cost in half. Cursor releases version 3.11, with Side Chats and dedicated thread search. On the platform side, GitHub makes Issue Fields generally available, and HeyGen continues its HyperFrames series with a seventh skill dedicated to AI-generated motion graphics.
Anthropic extends access to Fable 5 and Claude Code limits until July 19
July 12 — Anthropic is extending access to Claude Fable 5 across all paid plans until July 19, along with the weekly Claude Code limits raised by 50%. The mechanism remains the same as in the initial rollout: users can dedicate up to half of their weekly usage quota to Fable 5, then continue via paid credits or switch to another Claude model within the limit of their remaining quota.
“As before, you can use up to half of your weekly usage limit on Fable 5. After that, you can continue using Fable 5 with usage credits, or switch to another model to keep working within your remaining limits.” — @claudeai on X
This one-week extension follows the global rollout of Fable 5 on July 1 and the clarifications issued in early July regarding its cybersecurity safeguards.
🔗 @claudeai tweet — Fable 5 extension
Amp switches its default agent model from Opus to GPT-5.6 (~50% cheaper)
July 11 — Quinn Slack, CEO of Amp (Sourcegraph), announces on X that Amp has moved the majority of its agent calls from Claude Opus to GPT-5.6, for an average cost reduced by about 50% and response quality he says is better. The tweet, reposted by the official @AmpCode account, shows the effort dial “The Dial” (introduced on July 9) renamed “AMP · MK-IV SERIES”: the default configuration now places the agent on GPT-5.6 Sol X-High, and Fable 5 High as the “Oracle” model in charge of review and verification.
“Amp is now ~50% cheaper on average (and much better) since we mostly replaced Opus with GPT-5.6 and shipped this little guy. Wild times.” — @sqs on X
No dedicated post has been published on Amp’s official blog about this yet: the information currently comes only from X.
Cursor 3.11: Side Chats, conversation search, redesigned pickers
July 10 — Cursor publishes version 3.11 of its changelog, centered on three new features. Side Chats let you open a side conversation (/side, /btw, or the “+” button) that runs in parallel with the main thread without interrupting it, inherits its context, and can be recalled with an @-mention to bring its conclusions back in — by default oriented toward reading and clarification rather than code changes. Conversation search (Cmd+K in the Agents window, Cmd+F in the current thread) makes it possible to find an old thread beyond its name or PR number, via a local index sized for thousands of conversations. The project and repository pickers are redesigned (search limited to the work context, project creation with GitHub/GitLab/Azure DevOps connection without leaving the picker), and cloud agents gain new hooks (beforeSubmitPrompt, afterAgentResponse, stop, subagentStart) to observe and control their conversation more precisely.
GitHub Issue Fields reaches general availability
July 12 — GitHub announces the general availability (GA) of Issue Fields, a feature that lets you add structured, typed metadata — priority, estimated effort, dates, custom values — directly on issues, across all repositories in an organization. The stated goal: give a quick overview of a ticket’s status without opening each issue individually, a time saver for teams managing large volumes of tickets or running their work through GitHub Projects. The feature, previously in limited testing, is now open to all users and complements the platform’s native project management ecosystem (Projects, views, dashboards).
🔗 @github tweet — Issue Fields GA
HeyGen unveils the HyperFrames Motion-Graphics skill (day 7/30)
July 12 — In its daily presentation series for HyperFrames skills (30 days), HeyGen unveils the 7th episode: /motion-graphics, which generates short motion graphics (less than 10 seconds, with no narration) from a simple natural-language description — animated counters, charts, logo stings, text cards. The skill first determines whether it already has the necessary content (CSV, provided logo) or needs to search the web for it itself. HeyGen illustrates the concept with real examples: an animated chart powered by the actual HyperFrames npm download statistics (1,448,089 installs between June 11 and July 10), and a stats card featuring the repository’s 34,320 GitHub stars. The skill is available as open source, alongside the Talking-Head-Recut (day 6) already covered the previous day.
🔗 @HeyGen tweet — Motion-Graphics skill 🔗 GitHub — motion-graphics skill
What this means
Choosing the default model becomes a direct economic lever for makers of agentic tools. By switching its default agent from Opus to GPT-5.6, Amp reports an average cost cut in half without degrading — according to its own CEO — perceived quality: a sign that model competition is now playing out as much on cost/performance in real-world agentic usage as on raw benchmarks. Anthropic, for its part, extends Fable 5 access by one week and raises Claude Code limits by 50% — a way to keep users within its own ecosystem while third-party clients like Amp split some of their traffic in favor of a competitor.
Two announcements, each in their own way, structure the developer’s work around better-organized metadata and context. GitHub Issue Fields, now generally available, formalize priority, effort, or dates directly on issues rather than in ad hoc labels or third-party project management tools. Cursor 3.11’s Side Chats address a different but complementary need: carry out clarification in parallel with the main thread without polluting it, then reinject only the useful conclusions. In both cases, the challenge is the same — reduce noise and make information usable at scale, whether across hundreds of issues or thousands of conversations with an agent.
HeyGen methodically continues its daily demonstration of HyperFrames’ capabilities, its agent-driven video production environment. Today’s Motion-Graphics skill in day 7 illustrates a principle already seen with yesterday’s Talking-Head-Recut: the agent does not just execute a prompt, it decides for itself whether it should look up its data (npm stats, GitHub stars) or use what the user provided. As the series progresses, HeyGen is building less a catalog of isolated features than a library of reusable and combinable skills for fully automated video productions.