
Is Claude Opus 5.5 Nerfed? Claude Code Ignores Your effortLevel. How to Fix It
Opus 5.5 defaults to medium effort, and Claude Code 2.1.317 changed subagent effort on Nov 3. The settings fix, plus a script to test it yourself.

Opus 5.5 defaults to medium effort, and Claude Code 2.1.317 changed subagent effort on Nov 3. The settings fix, plus a script to test it yourself.

OpenAI's own report: Astra hit ~10% of rubric items on safety tasks vs ~16% otherwise, and its monitors didn't flag it. Six checks that catch it.

One 128 GB DGX Spark won't hold the 594B weights due next week, and DeepSeek's 552B model, out now, beats it on eight of the nine coding benchmarks they share.

That 290MB is one cropped Task Manager screenshot, so measure your own Discord client, change the settings worth changing and skip Ripcord.

93% of base Qwen3.5-0.8B's MATH-500 answers never finished. One HN commenter says thinking off alone gets 52.0%. Plus how to run the edit locally.

Blocks, likes and follows sit in your public AT Protocol repo. The curl commands to check yours, and why a mute stays private when a block doesn't.

Devs who learned with AI scored 17 points lower, worst on debugging. The ones who kept pasting errors back into the AI did worse still: about 24%.

Published Pi 4 benchmarks show which small LLMs the Zero 3W could run, and whether you should buy a $15 Zero 2W now or wait.

CPython's JIT leaves main early in the 3.16 cycle. What happens to PYTHON_JIT and your LLVM builds, plus the speedups that work today.

It also went for the 169.254.169.254 metadata IP and used unshare to get root in a namespace. Four settings to change before a local agent runs tools.

The weights land Oct 31 with no named license. The Q4/Q2 memory math, which Macs and servers can load it, and the llama.cpp offload command.

php -v can say you're fixed while Apache isn't, and Ubuntu's patched PHP is still versioned 8.3.6. The checks that show what your web server really runs.

Per token it's 50% cheaper than M3, not 70%, and it only undercuts DeepSeek V4.1 Flash during DeepSeek's peak hours.

The NSA's mystery P-256 seed is lrand48 output you can reproduce in 11 lines of C. Plus: should you still pick Ed25519 for SSH, JWT and TLS?

Eight to ten features were reverted, Postgres 14 hits EOL a week before GA, and here are the pg_upgrade --check steps to run against 19 RC1 now.

Linux is dropping the Assisted-by tag while QGIS bans AI code. Torvalds wrote patches with Claude Code, and his rule fits in your CONTRIBUTING.md.

Mozilla hasn't published a price for its new console, while Chrome Enterprise Core costs $0. Here's what regular Firefox users keep for free.

It's an embedding model that was never trained to transcribe. How it stacks up against Whisper and Parakeet on WER, license and the Mac-only catch.

Apple and Google now let you export passkeys, so lock-in is mostly solved. One phone holding all of them isn't. Here's the two-key backup setup.

About 40 of ~50 AI-tagged QGIS PRs were closed before the ban. How Gentoo, QEMU, Godot, LLVM and the Linux kernel handle AI code.

A minute of video costs about a quarter of a cent on Workers AI, and Cloudflare's own benchmarks show the jobs where Jev still wins.

It leads OmniDocBench by 0.05 points at 4x the size, its license excludes the EU, and the default vLLM setup realistically wants a 24 GB GPU.

670 MB in Q8_0, one llama-server command, and a moderation score that beats Decider 2B. Plus the license clause the launch post skips.

d1-3B reads images but only returns probabilities, never text. llama.cpp and transformers setup, a 1.67 GB GGUF, and the $10M license catch.

Tahoe still holds UNIX 03 and macOS 27 isn't registered. Plus why the badge never stopped sed -i '' from breaking your Linux CI, and the portable fix.

Glyph and Fernando Borretti say AI is draining something human out of programming, and HN pushed back. Plus a git one-liner that shows how alone you ship.

The full pipeline is 16.2B params and 32.4 GB in BF16. Quantized weights down to 11 GB, the ComfyUI sigma fix for mushy output, and a non-commercial license.

Deno Deploy shuts down in six months and the announcement never mentions Deno KV. What's confirmed, what isn't, and the grep to run this week.
If you've been anywhere near the Python ecosystem in the last year, you've probably felt the ground shift under your feet. Tools like uv and Ruff — built by Astral — went from "interesting experiments" to "how did I ever live without these" at breakn...