Anthropic's new flagship model matches its priciest rival while costing far less to run.
What the article says
- Opus 5.5 now performs about as well as Anthropic's top tier model but runs much cheaper and faster.
- It handles huge coding jobs well, auditing a massive codebase in hours instead of days.
- Anthropic says the model writes more clearly and puts key information first, addressing a common complaint about earlier releases.
- It ships with tighter safety checks for biology and cybersecurity work, routing sensitive requests to a more limited model.
- Subscribers get higher usage limits and a saved rate limit reset they can use whenever they want.
What HN is saying
- Some read the launch as clashing with Anthropic's pledge last week to slow down releases, pointing at the aggressive price cuts and performance claims as proof they haven't actually paced anything.
- Opinions split on the writing style. A few say it finally reads clearly, others say it is still wordy and leans on the same pet phrases as before.
- Multiple people say OpenAI's rival model still feels faster and less bloated, even though this one costs far less per token.
- Several worry the new biology and cybersecurity restrictions will just push serious researchers and hackers toward less restricted Chinese models.
- A few commenters said cheap open models already do everything they need for a fraction of the cost.
OpenAI cuts prices on its coding models by half and closes the gap with Claude.
What the article says
- OpenAI released two cheaper models, Sol and Luna, sitting below its flagship Astra model.
- Both cost half as much as their predecessors while matching or beating them on coding and business benchmarks.
- On coding tests, Sol nearly matches Anthropic's top models at a fraction of the price.
- Prompt caching got better too, cutting the cost of long running agent sessions.
- OpenAI also says both models communicate more clearly, with less jargon and fewer odd phrasings.
What HN is saying
- Commenters mostly cheered the price cuts, calling Luna an absurdly cheap and capable model.
- Some coders are attached to the old Sol model's style and worry the new one will feel different to work with.
- A running argument compared Claude Code against Codex on usage limits and context window size, with no clear winner.
- Independent benchmarks show a split result, Sol improved but Luna actually scored slightly worse than its predecessor on coding tests.
- A few readers mocked OpenAI's confusing model names, joking that pricing is the only way to tell them apart.
Why AI written pull requests and design docs feel exhausting, and the one place AI writing actually helps.
What the article says
- The writer is fed up reading design docs, pull request summaries, and messages written entirely by AI, calling them detailed but hollow and stripped of context.
- Writing works because it carries voice and relationship, not just facts, so a machine's summary loses the meaning that mattered.
- For a paper he wrote himself, he found AI genuinely useful for catching errors, checking claims against the code, fixing grammar, and drawing diagrams.
- The one section he let AI write in full was the abstract, since that's the most mechanical, least human part of any document.
- He argues good writing needs uncertainty and lived experience, things a model optimizing for the next word simply can't supply.
What HN is saying
- Commenters agree writing transfers a feeling or experience, not raw information, so AI padding out a short thought just adds noise to filter.
- Several say they now reject pull requests or messages that feel machine bloated, since reading them takes longer than doing the work would.
- The sharpest pushback argues it's fine to use AI if you proofread and iterate until the result genuinely sounds like your own voice.
- One engineer said their team already skips human review entirely, letting AI write and review pull requests before automatic merging.
- A few noted the essay's own opening line reads like AI prose, sparking a side debate over whether that was deliberate irony.
A blogger turned off Apple Intelligence, upgraded macOS, and found Apple had quietly switched it back on.
What the article says
- He disabled Apple Intelligence and Siri on an older macOS, but upgrading to the newest version silently turned everything back on.
- Turning Siri off again still leaves several background processes running that keep eating memory and writing data.
- Screen Time's parental controls hide some AI menus but don't actually switch the features off.
- The AI download eats over twenty gigabytes of disk space, real money on Apple's pricey storage tiers.
- He argues the AI industry treats consent as optional and keeps forcing itself back on regardless.
What HN is saying
- Several commenters push back on the phoning home claim, arguing the report toggle only shows existing data rather than a new leak.
- Many agree Apple's opt out prompts never really let you say no, they just come back with each update.
- People swap workarounds like faking your region to Russia or the EU, or digging into Screen Time restrictions, to properly kill the AI features.
- A parallel debate breaks out over Linux versus Mac, with some praising Apple's hardware and others saying no operating system fully respects you.
Bryan Cantrill says Sun didn't fail from bad tech. It got bored with running a business.
What the article says
- Cantrill argues Sun's real failure was losing interest in the basics of selling and serving customers, not its engineering.
- A 2005 story sums it up. A fast growing startup running on Solaris tried to buy a pile of Sun hardware and couldn't get anyone at Sun to pick up the phone.
- Dell, by contrast, called back overnight and closed the whole deal fast, making the customer feel like a huge account.
- That startup later hired the Dell rep, Steve. Years afterward, Cantrill and Steve went on to co-found Oxide together.
- Cantrill's takeaway is that a company which stops caring about the mechanics of running a business can't survive, no matter how good its strategy is.
What HN is saying
- Commenters swap old war stories about how painful buying from Sun or DEC felt next to just ordering from Dell online.
- Many blame specific missteps, killing x86 Solaris early, passing on a deal with Google, and clinging to SPARC long after x86 caught up.
- Several argue the real killer wasn't any one decision but cheap x86 boxes running Linux simply getting good enough.
- A recurring gripe is that Sun's sales commissions rewarded contract size rather than profit, so reps allegedly sold deals that lost money just to get paid.
- One thread argues Sun's deeper mistake was never taking user interface seriously the way Apple did.