September 2, 2026·5 min read
Firefox 155 speeds up page loads
Firefox 155 goes faster, Meta rivals Claude Code, and more.
Hi there, this is your daily ☕️ Devshot.
In today's Devshot:
🦊 Firefox 155 speeds up page loads
💻 Apple allows dropping Intel Mac support
⚔️ Meta launches Claude Code rival
🤖 Claude Fable 5.1 now 45% cheaper
🗜️ Cloudflare tests Zstandard compression
Plus: 🎁 8 other news you might like, 🧰 6 tools, and 📚 5 papers.
🦊 Firefox 155 speeds up page loads LINK
Firefox 155 binaries are out ahead of the formal announcement on Tuesday, headlined by faster page loads through Happy Eyeballs v3, which races multiple connection methods at once and keeps whichever succeeds first.
Happy Eyeballs v3 extends parallel IPv4/IPv6 selection to read modern DNS records for the protocols a server supports, concurrently attempting HTTP/2 and HTTP/3 to settle on the fastest available connection.
The release also adds QUIC version 2 for HTTP/3 whenever a server negotiates it, introduces an optional AI-powered Smart Window in the US, Canada, and France, and fixes a Linux sleep bug after long sessions.
💻 Apple allows dropping Intel Mac support LINK
Apple has told developers they no longer need to ship Intel-compatible apps, and confirmed that Intel-only apps will stop running entirely once macOS 27 arrives.
macOS Tahoe is the last release supporting Intel Macs and Rosetta-translated Intel apps, so the Mac App Store now accepts Apple Silicon-only submissions rather than rejecting non-universal builds.
The wind-down runs in stages, starting with macOS 26.4 prompting users to update apps, and Apple is urging developers who haven't yet moved to Apple Silicon to begin immediately.
⚔️ Meta launches Claude Code rival LINK
Meta has taken Muse Code, its terminal-based coding agent and answer to Claude Code and Codex, out of beta less than a month after launch, adding new features and three subscription tiers running $5 to $50 monthly.
The release adds inter-session messaging to share context between overlapping sessions, Workflows for orchestrating specialized agents with a control room, and a rewind feature that rolls conversation and code changes back to an earlier point.
A new SDK in developer preview, powered by the Muse Session Protocol, lets you drive Muse Code programmatically, starting, resuming, and steering sessions from your own apps, IDE integrations, or agents via a TypeScript SDK or direct MSP implementation.
🤖 Claude Fable 5.1 now 45% cheaper LINK
Anthropic shipped Claude Fable 5.1, cutting the price 45% while beating its predecessor and Opus 5 across every published benchmark, though it still costs twice as much per token as Opus 5 at $10 input and $50 output.
Fable 5.1 hit 52.6% on Terminal-Bench-Science 0.1 versus Fable 5's 24.7%, and 55.8% on agentic-coding benchmark Terminal-Bench 4.0, ahead of Opus 5's 52.3%, with a restricted Mythos 5.1 sibling gated to vetted cybersecurity and life-sciences users.
Available now via the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry under model ID "claude-fable-5-1," it draws pay-as-you-go credits on Pro and standard seats but ships included for up to 50% of weekly usage on Max plans.
🗜️ Cloudflare tests Zstandard compression LINK
Cloudflare is prototyping Cache Transcoding, encoding eligible cached responses with Zstandard inside its Pingora proxy to fit more customer content on the same hardware while trading a few percent of extra CPU.
Compressible text like HTML, JSON, CSS, and JavaScript at or above 4 KiB gets encoded at zstd level 3 on a cache miss, shrinking to roughly a third of its size, then decoded before reaching clients.
The compressed form stays intact across Tiered Cache hops, cutting cross-data-center backbone traffic, and a storage marker stops re-encoding; testing over one million requests confirmed the ~2.8x compression held on a deliberately compressible corpus.
Other news you might like
- Python 3.15.0 candidate 2 is here!LINK
- Kubernetes v1.37: etcd RangeStream Cuts Memory Use on Large List ReadsLINK
- Developer uses Claude Code to debloat Android smart TV for 'unbelievable' performance upgrade — TV now smoother than it was new as autonomous agent deactivates apps, shortens animations, all without root accessLINK
- Meta launches Muse Voice Transcribe for real-time voice dictation on MacLINK
- Google Gemini's new agent-based video analysis cuts token usage by up to 88 percentLINK
- I Vibecoded Morrowind, Duke Nukem, and Serious Sam Onto an iPhoneLINK
- HashiCorp Vault agentic IAM is now generally availableLINK
🧰 Trending tools
AnySearch: queries multiple trusted sources in parallel for AI agents, returning filtered, de-duplicated results as structured data ready to useLINK
SKI: adds voice control to AI coding agents like Claude Code and Codex, letting you dictate commands and hear responses hands-free while codingLINK
Coldtea.ai: automates QA and monitoring using AI agents that catch post-deployment bugs, flag issues, and generate actionable fix tasks for your teamLINK
ZooData: converts webpages into structured JSON to cut LLM token usage by 75%, with pre-analyzed Amazon and TikTok e-commerce data for AI agents.LINK
Kane CLI: run browser tests from plain-English commands in real Chrome, no selectors or brittle scripts, ideal for developers, QA engineers, and AI agentsLINK
Openbase: helps developers pick npm and open-source packages by comparing popularity, maintenance activity, and reliability metrics alongside real user reviews.LINK
📚 Trending papers & reports
Quantum software testing remains a fragmented field where academic and industry sources use scattered tools and methods, held back by shaky hardware, unpredictable results, and immature tooling that make quality assurance hard.LINK
Coding agent teardown dissects the source code of eleven popular AI programming tools, finding they all rely on hand-built control loops rather than off-the-shelf frameworks or the vector-search approach most assume powers modern AI.LINK
AI tool failures rarely tell software what to actually do next, since across 21 tested cases error messages flagged that something broke but almost never gave a machine-readable fix, forcing agents to guess from plain text.LINK
Deployment safety records stay truthful about what software is actually running even when an update crashes halfway, hitting 100% agreement across 1,200 failure tests while recovering in under half a second.LINK
Complex AI coding tools get beaten by a single prompt on a newer model in 37% to 63% of recent research papers, showing much engineering just patches temporary model weaknesses that vanish with the next release.LINK
See you tomorrow for a new dose of ☕️ Devshot!
