--- title: "Opus 5 Shipped and Hacker News Argued About a Computer Vision Pipeline" pubDatetime: 2026-07-27T23:33:00.000Z description: "Claude Opus 5 arrived at Opus 4.8 prices with a #1 Intelligence Index ranking, and the 1,300-comment HN thread mostly skipped the benchmarks to argue about one demo anecdote. The pricing story has a documented counterexample." tags: [anthropic, claude-opus-5, benchmarks, pricing, community, 2026, 2026-q3, 2026-07] --- > [!tldr] TL;DR > Anthropic [shipped Claude Opus 5](https://www.anthropic.com/news/claude-opus-5) on Friday 24 July at $5/$25 per million input/output tokens, the same rate as Opus 4.8 and half of Fable 5, and [Artificial Analysis](https://artificialanalysis.ai/models/claude-opus-5) put it first out of 190 models on its Intelligence Index the same day. The [HN thread](https://news.ycombinator.com/item?id=49038433) has run past 1,300 comments, largely skipping the benchmark table to argue about one launch anecdote: Opus 5 wrote its own computer vision pipeline to reconstruct a part it couldn't see. The most useful lesson in the thread is about pricing, not capability. Per-token price says little about per-task cost, and [aipricing.guru measured](https://www.aipricing.guru/news/anthropic-claude-opus-5-pricing-impact-july-2026/) Opus 5 spending 1,989 completion tokens on a narrow task where Opus 4.8 spent 630. Run your own comparison on your own workload before you migrate anything. Anthropic put that anecdote in the launch post as evidence of initiative. Hacker News read it three ways: as a milestone, as a billing problem, and as pattern-matching from training data. Only the first is the reaction a launch page hopes for. ## The demo that started the argument The first reaction is the one Anthropic presumably wanted. "How surreal is it that we are not absolutely jaw-dropped by these types of capability improvements?" [wrote makaking](https://news.ycombinator.com/item?id=49045107). "It's been less than 4 years since ChatGpt came out and now they are spontaneously building their own ML pipelines to do real-world 3D modeling tasks reliably." The sharpest reply reframes the same behaviour as a product bug. "While this is certainly impressive, I believe it is not a good answer from a user's perspective," [wrote carra](https://news.ycombinator.com/item?id=49045729). "Instead of having Claude automatically spend tokens in an unrelated task, I would have preferred it to first tell me: 'I can't view the drawing. Would you like me to build a computer vision pipeline to do it'?" A model that decides mid-task to build infrastructure is spending your money on a plan you never approved. Anthropic sells that initiative as thoughtfulness. Anyone paying by the token gets to decide whether they agree. A third strand questions the word "own". "Was it 'its own' or something that was part of its training material?" [asked teiferer](https://news.ycombinator.com/item?id=49045364), who also said the technology makes their work "10x easier and quicker" before adding: "But it's not like it's inventing this stuff from scratch / first principles. It has seen this kind of tech before by consuming all publicly available source code and books etc." A separate branch of the thread drops the capability question entirely and worries about [job displacement](https://news.ycombinator.com/item?id=49046662). For a launch-day thread on a frontier model, the mood is more philosophical than triumphant. ## The efficiency claim has a documented counterexample Close to Fable 5 at half the price is a per-token claim, and per-token claims survive only until someone measures tokens per task. [aipricing.guru ran that test](https://www.aipricing.guru/news/anthropic-claude-opus-5-pricing-impact-july-2026/) and found Opus 5 generating 1,989 completion tokens against 630 for Opus 4.8 on a narrow task. Same price per token, roughly three times the output. On that task, the cheaper model was the old one. One narrow test settles nothing. Those extra tokens could be buying a better answer on that particular task. The point is that Anthropic's launch materials don't show the comparison at all, and one third party who ran it got a result pointing the other way. If your workload is high-volume and shallow, benchmark it before you switch the default. The vendor benchmark claims deserve the same treatment. Artificial Analysis independently confirms the Intelligence Index ranking, but the eye-catching numbers in the announcement (ARC-AGI 3 at roughly three times the next-best score, Zapier AutomationBench at 1.5x) are Anthropic measuring Anthropic. No third-party eval has corroborated them yet. ## Fable 5 lost 18 days to an export-control order The loudest community framing is that Fable 5 now looks pointless: if Opus 5 gets close for half the money, what was the drama for? That reading skips what actually happened. On 12 June the US Commerce Department [ordered Anthropic to suspend](https://www.anthropic.com/news/fable-mythos-access) all access to Fable 5 and Mythos 5 over a demonstrated jailbreak with claimed national-security implications, three days after Fable 5 shipped. Anthropic said it disagreed with the suspension. It did not dispute that a real vulnerability existed. Access came back on 30 June, [after a letter from Commerce Secretary Lutnick](https://www.cnbc.com/2026/06/30/anthropic-says-trump-admin-has-lifted-export-controls-on-claude-fable-5-and-mythos-5.html) lifted the requirement, 18 days later. So Anthropic spent 18 days with its flagship line blocked by a federal order, and six weeks after that order it shipped a cheaper model that reaches most of the same capability and carries none of that baggage. Read as marketing choreography, the sequence is incoherent. Read as a company that wants a strong product it can be confident of keeping switched on, it makes sense. ## Where Opus sits in the lineup now Opus used to be the flagship name. It is now the default on Claude Max and the strongest option on Claude Pro. Reading the price list as a three-tier lineup, Mythos and Fable above, Opus as the value option, Haiku underneath, is an interpretation of pricing rather than a company position. Anthropic has published no statement describing a restructured tier strategy. The question of why anyone would still pay for Fable has a duller answer than the thread allows. Anthropic's own wording is that Opus 5 comes close to Fable 5, not that it passes it, and the independent ranking that put Opus 5 first out of 190 models doesn't score it against Fable 5 by name. Close at half the price is a good deal. It is still close. What is solid: Opus 5 tops an independent intelligence ranking, costs what Opus 4.8 cost, and is available now, give or take the [elevated errors](https://news.ycombinator.com/item?id=49068029) users reported in a smaller HN thread on 27 July. See also [what Opus 5 is allowed to find](/posts/claude-opus-5-find-vulns-not-exploit) on its security posture. Read the system card, then read your token bills. --- **Sources** - [Introducing Claude Opus 5 (Anthropic, 24 July 2026)](https://www.anthropic.com/news/claude-opus-5) - [Claude Opus 5 (Hacker News, 1,300+ comments)](https://news.ycombinator.com/item?id=49038433) - [Claude Opus 5 (Artificial Analysis)](https://artificialanalysis.ai/models/claude-opus-5) - [Simon Willison, "Introducing Claude Opus 5"](https://simonwillison.net/2026/Jul/24/introducing-claude-opus-5/) - [Anthropic statement on the US directive to suspend Fable 5 and Mythos 5](https://www.anthropic.com/news/fable-mythos-access) - [CNBC, "Anthropic says Trump admin has lifted export controls"](https://www.cnbc.com/2026/06/30/anthropic-says-trump-admin-has-lifted-export-controls-on-claude-fable-5-and-mythos-5.html) - [aipricing.guru, Claude Opus 5 pricing impact](https://www.aipricing.guru/news/anthropic-claude-opus-5-pricing-impact-july-2026/) - [TechCrunch, "Anthropic launches Opus 5"](https://techcrunch.com/2026/07/24/anthropic-launches-opus-5/) **Related on this blog** - [Opus 5 Is Allowed to Find Bugs Now](/posts/claude-opus-5-find-vulns-not-exploit) - [The White House Says Kimi K3 Is Distilled Fable](/posts/moonshot-fable-distillation-kimi-k3)