☕️ Anthropic releases Claude Opus 5

Hi there, this is your daily ☕️ Devshot.

In today's Devshot:

🤖 Anthropic releases Claude Opus 5

🐧 AMD adds low-power CPU cores to Linux

⚖️ Zalando's load balancer hits 1M req/s

☸️ Amazon EKS adds Kubernetes rollback

📊 TanStack Table V9 beta cuts memory use

🦀 Ruff v0.16.0 ships

Plus: 🎁 13 other news you might like, 🧰 6 tools, and 📚 5 papers.

🤖 Anthropic releases Claude Opus 5 LINK

Anthropic has shipped Claude Opus 5, now the default model on Claude Max and the top-tier option on Claude Pro, promising better efficiency on software engineering and knowledge work while holding prices steady.

Pricing stays at $5 per million input tokens and $25 per million output tokens, with a Fast mode running roughly 2.5x quicker at double the base rate, accessed through the claude-opus-5 API endpoint.

The launch bundles two betas: mid-conversation tool changes on the Claude Platform, and automatic API fallbacks that reroute requests blocked by safety classifiers to alternative models, with no data retention required for general access.

🐧 AMD adds low-power CPU cores to Linux LINK

AMD's Zen 5 core in Turin can sustain 58 concurrent cache-line fetches from a single core, roughly double Intel's Granite Rapids at 30, giving it the clear lead in memory-level parallelism.

Measured with a pointer-chase benchmark over a 1 GiB array, the m8a instance's single core hits 24.5 GiB/s of random-access throughput, about twice Intel's, though its 142 ns latency is slightly worse than Granite Rapids' 133 ns.

AMD's concurrency climbed from 15 on Naples in 2018 to 22 on Milan and then jumped to 58 on Turin, while Graviton 5 stalled at 19 but delivered the only latency improvement in the collection at 96 ns.

⚖️ Zalando's load balancer hits 1M req/s LINK

Zalando's engineering team built an in-process, client-side load balancer for its Product Read API, handling roughly 1 million requests per second with more predictable latency, lower infrastructure costs, and clearer visibility into failure origins.

The balancer reimplemented Skipper's exact algorithm using xxHash64 with 100 virtual nodes per endpoint so both routing paths hit identical pods during migration, while a watch-based Kubernetes informer replaced control-plane-crashing polling.

Ramping traffic from 1% to 100% behind toggles shrank Skipper's fleet from 50+ pods to 8 and cut daily deployment cost from $450 to $110, with N-ring fade-in over 30 seconds warming new pods gradually.

☸️ Amazon EKS adds Kubernetes rollback LINK

Amazon EKS now lets you roll back a cluster's control plane to its previous Kubernetes version within 7 days of an upgrade, giving teams a way to recover quickly from problematic in-place cluster updates.

Rollbacks revert the API server and control plane one minor version at a time while preserving data, workloads, and persistent volumes, and EKS runs cluster insights beforehand to flag node version mismatches or add-on dependencies.

Control plane rollbacks work on all EKS clusters, but node rollbacks are limited to EKS Auto Mode, which reverts worker nodes first and offers a cancel API to stop, accelerate, or redirect the process.

📊 TanStack Table V9 beta cuts memory use LINK

TanStack Table V9 has entered beta, a foundational rewrite that overhauls state management, shrinks memory footprint at scale, and reduces bundle size while preserving the core table logic developers already use across React, Vue, Angular, Solid, Svelte, Preact, and Lit.

Features are now opt-in rather than bundled: V9 registers only what a table uses, so a small table starts around 5kb while enterprise grids pull in sorting, filtering, and pagination on demand, with TypeScript enforcing that each API exists only once its feature is registered.

State now flows through TanStack Store atoms and per-slice subscriptions instead of the old model, enabling fine-grained re-renders; maintainer Kevin Van Cott says the stable React Compiler forced the rewrite because V8's patterns broke under it, and migration hooks ease gradual upgrades.

🦀 Ruff v0.16.0 ships LINK

Ruff v0.16.0 shipped on July 23rd, expanding its default rule set from 59 to 413 enabled rules, so unpinned installations began surfacing new violations and breaking CI jobs without any configuration changes.

Ruff now bundles 968 total rules, up from 708, many catching syntax errors and immediate runtime failures previously left off by default; try them via `uvx ruff@latest check .` on any project.

Running `uvx ruff@latest check . --fix --unsafe-fixes` against sqlite-utils flagged 1,618 errors and auto-fixed 1,538, leaving issues like naive `datetime.now()` calls, blind `except Exception`, and useless attribute access with clear explanations.

Other news you might like

• How routing keys isolate Kafka consumer tests on a shared brokerLINK

• CSS Infinity Use CasesLINK

• A Beginner’s Guide to Clocks, Causality, and Ordering in Distributed SystemsLINK

• How to clean up AI-generated code with FallowLINK

• How I Find Problems to Solve as a Staff EngineerLINK

• Netflix Details Its In-House LLM Serving Platform with Triton and vLLMLINK

• HTTP Message Signatures with curlLINK

• Google Chrome finally gives Arm64 Linux users the full browser experienceLINK

• Devin Desktop Review: Windsurf Becomes an Agent HubLINK

• Five new Rust toolsLINK

• Microsoft explains why its West US Azure and cloud services failedLINK

• Cursor's agent swarm suggests cheaper models can handle most coding when frontier models plan the workLINK

🧰 Trending tools

ditto.site: turns any public URL into clean Next.js or Vite code deterministically, preserving components, design tokens, and hover states via a free API or MCP server.LINK

Openbase: helps developers pick reliable open-source packages by comparing popularity, activity, reliability metrics, and user reviews side-by-side.LINK

Pushary: delivers AI agent approval requests to your phone's lock screen, letting you approve or deny actions remotely while automating routine decisions with configurable policies.LINK

Second Brain for AI v2: a self-hosted memory layer that lets Claude, ChatGPT, and Cursor share persistent context via semantic search, eliminating repetitive re-explanations across sessions.LINK

ShellMate: a cross-platform SSH client with collaborative workspace features, letting dev and DevOps teams manage remote servers and share sessions together.LINK

Tamamon: a macOS desktop pet that evolves based on your Claude Code usage, tracking sessions locally to feed and grow 20 collectible species.LINK

📚 Trending papers & reports

AI-written C code for embedded devices gets automatically checked and repaired, cutting security flaws from 49% to 19% and slashing total code errors by 83.7% in one tested model.LINK

Code-as-images billing shows that turning source code into pictures instead of text can cut the "input tokens" providers charge for by up to 86.5%, but Google's system sometimes charges nearly 7 times more for small files, so the savings depend entirely on which API you use.LINK

AI-assisted coding practice called vibe coding speeds up prototyping in 45% of 47 reviewed sources, but evidence on long-term code quality and safety remains thin, so treat it cautiously beyond early-stage builds.LINK

Coding agent instructions should match how each model outputs edits, since Doubao succeeded 94% of the time using JSON Patch while DeepSeek hit 66% with unified diff, meaning one-size-fits-all formatting quietly wastes an agent's real ability.LINK

AI code review pairings show having Claude check Codex's draft code lifts its pass rate from 71.6% to 89.7%, while having Codex review Claude's work actually drops accuracy from 91.4% to 82.8%.LINK

See you tomorrow for a new dose of ☕️ Devshot!