|
|
Hi there, this is your daily ☕️ DevShot.
|
|
|
In today's DevShot:
|
|
🐋 DeepSeek V4.1 Flash beats bigger rivals 🐧 Linus Torvalds releases Linux 7.3 🐍 Google adopts Pyrefly for Python 🐈 NestJS 12 adds native observability ⚡ Zod compiles schemas 9x lighter in memory Plus: 🎁 9 more stories you might like, 🛠️ 5 tutorials, 🧰 7 tools, and 📚 5 papers.
|
|
Discover laptops with built-in intelligence.Shift your workforce into hyper-productivity mode with Dell Pro laptops powered by Intel® Core™ Ultra with Intel vPro®. Shop on Dell Premier.
|
🐋 DeepSeek V4.1 Flash beats bigger rivals
LINK
|
- DeepSeek shipped V4.1-Flash-1017, a smaller model that tops its predecessor and larger rivals like Claude Opus 5 and GPT 5.6-Sol on several coding and agent benchmarks, with weights open-sourced on Hugging Face and ModelScope.
- On the DeepSWE benchmark it scores 77% at just under $0.20 per task, beating the prior version's lower accuracy at over twice the cost, thanks to shorter outputs and training that uses the model's own confidence as a reward signal.
- DeepSeek also opened its Search & Fetch APIs to all developers at $0.75 per 1,000 calls, free through server-side tools in DeepSeek Harness, Claude Code, and Codex until October 31, though Beijing reportedly plans to tighten open-weight release rules.
|
🐧 Linus Torvalds releases Linux 7.3
LINK
|
- Linus Torvalds has shipped the final Linux 7.3 kernel on schedule, with the 7.4 merge window opening tomorrow, though he warns merges will run on an unusual timetable while he's on annual family vacation.
- The release set records with 18,765 non-merge commits from 3,021 developers, 807 of them contributing to the kernel for the first time.
- Notable additions include the fchroot() system call, global per-CPU data for BPF programs, completion of the sched_ext subscheduler feature, working-set tracking for VMs, binfmt_misc extensions, and a new failfs filesystem that fails every operation.
|
🐍 Google adopts Pyrefly for Python
LINK
|
- Google has dropped Pytype, its long-running Python type checker, in favor of Pyrefly, the open-source checker Meta released last year, citing faster, more accurate checks and lower infrastructure cost.
- Google's internal benchmarks put Pyrefly at roughly 10x the speed of Pytype, while cutting incremental rebuild latency by up to 98% and roughly halving the peak compute cost for type checking.
- Written in Rust, Pyrefly passes about 97% of the Python typing conformance suite and checks 1.8 million lines per second, fast enough to type-check on every keystroke, with inline diagnostics that pinpoint error causes and sometimes suggest fixes.
|
🐈 NestJS 12 adds native observability
LINK
|
- NestJS 12 shipped as a major release, moving its core packages to ESM, adding first-class Standard Schema validation, rebuilding the CLI, and introducing an official observability SDK that reports traces through the framework's own lifecycle.
- The ESM switch requires Node.js 20.19+ or 22.12+ but keeps CommonJS apps working, and preview benchmarks found no real startup or memory gains, framing it instead as a cleanup of tooling and configuration debt.
- Route decorators now accept any Standard Schema object, so Zod, Valibot and ArkType work without wrappers, though types still need manual annotation and critics warn the NestJS-only observability SDK risks framework-level vendor lock-in.
|
⚡ Zod compiles schemas 9x lighter in memory
LINK
|
- Zod 4.5, released on August 28, 2026, added in-process schema compilation and cut schema memory usage by roughly nine-fold, with a follow-up 4.6 release on September 9 extending the work and fixing a recursive-schema memory regression.
- Compilation turns any schema into a precompiled validator that parses three to nine times faster, runs lazily at first parse with no bundler plugin or codegen, while memoising bound methods drops a bare schema from 7.5kb of retained heap to 784 bytes.
- Compiled Zod hits 47.5M ops/sec on the Moltar parseSafe benchmark, edging past Typia and leaving Valibot and ArkType far behind, though the fastest validator stays workload-dependent and the 4.5 regression could cause out-of-memory failures until you upgrade to 4.6.
|
|
State of Product reveals what AI still hasn’t solved80% of product professionals say AI helps them ship faster, but customers aren’t seeing value any sooner. Atlassian’s State of Product 2027 explores the gains AI is delivering and the challenges that remain, from decision-making that hasn’t kept pace to gut instinct overriding customer evidence.
📚 Tutorials & Advice
|
> What requirements don't you have?: Identify which requirements a system can safely relax, since dropping needs like fresh data or concurrency yields simpler, faster designs that coding AI tends to over-engineer.
|
|
|
🔗 Other News & Articles
-
Introducing Scout 1.0: A new SRE model powering Bits AI with frontier performance at a fraction of the cost
LINK
-
How Datadog defines service criticality at scale
LINK
-
Introducing Claude Platform Tracing in New Relic AI Monitoring
LINK
-
His scanner wouldn’t work on Linux — so he let Claude study the Windows driver
LINK
-
Decoding base64 in Python at 10 gigabytes per second
LINK
-
Instacart cut nodes 70% and latency 50% with Dragonfly
LINK
-
Developer demonstrates PS5 emulation on Xbox Series X
LINK
-
Replit is no longer working with Anthropic, CEO says
LINK
-
One prompt, dozens of API calls: rebuilding the monday.com API edge for AI Agents
LINK
|
|
📄 Papers & Resources
|
> Spin-qubit control software gives both hardware engineers and algorithm researchers a single customizable toolkit to tune, run, and benchmark experimental quantum chips, letting teams prototype new quantum devices and test procedures without rebuilding their control stack.
LINK
|
|
> Enterprise API training data can be generated straight from an API's written spec without ever calling the live system, letting companies safely train tool-using AI agents on sensitive APIs that are too risky or costly to run.
LINK
|
|
> Hardware glitches can make AI models give wrong answers on math and coding tasks, showing that chip-level errors, not just software, are a source of unreliable output worth tracking.
LINK
|
|
> Fortran-to-Python translation turns an AI's one-off conversion choices into reusable, mathematically checked rules, producing Python that matches the original scientific code exactly on every tested case, with over 90% of rules reused across 14 of 15 programs.
LINK
|
|
> SIEVE is an open-source tool that lets software researchers build custom code datasets by pulling individual functions and classes from GitHub, each tagged with quality metrics and a score estimating whether it was AI-written.
LINK
|
|
|
|
🧰 Latest Tools
| ElevenLabs: turns any script into studio-quality speech in seconds, with 5,000+ voices across 70+ languages. Free to start. GENERATE FOR FREE |
|
Die With Me: a Mac app that tracks Claude and Codex usage, compares it with friends, and opens a chatroom when your allowance runs out.
LINK
|
|
Kilo Code for JetBrains: open-source AI coding agent for JetBrains IDEs that lets you build and iterate with flexible model choices across projects.
LINK
|
|
Mastra Factory: a TypeScript framework for building AI agents with staged workflows, memory, streaming, evals, tracing, and a Studio UI for testing.
LINK
|
|
Kilo Code: open-source mobile AI coding agent that pushes notifications when sessions need input, letting you dictate prompts and attach files remotely.
LINK
|
|
Inferock: a local proxy for OpenAI, Anthropic, Gemini, and OpenRouter calls that tracks token usage, failures, and retries to reveal billing overcharges.
LINK
|
|
Enter Pro: an AI-native platform for building and scaling apps, combining development and growth so you avoid stitching together separate tools and services.
LINK
|
|
|
|
💬 How did you find today's edition?
We read every reply — just reply to this email and let us know how we can improve!
|
|
|
