<p>I didn't have early access to today's <a href="https://www.anthropic.com/news/claude-fable-5-mythos-5">Claude Fable 5</a> release, but I've spent the past ~5.5 hours putting it through its paces. My initial impressions are that this is something of a <em>beast</em>. It's slow, expensive and has been quite happily churning through everything I've thrown at it so far. As is frequently the case with current frontier models the challenge is finding tasks that it can't do.</p> <p>First, let's review the key characteristics.</p> <p>Anthropic claim that <a href="https://www.anthropic.com/news/claude-fable-5-mythos-5">Claude Fable 5</a> offers the same performance as Claude Mythos 5, except with much more strict guardrails in place to prevent it being used for harmful things. Those guardrails trigger often enough that the Claude API has new mechanisms for letting you know when you hit them, and even has a <a href="https://platform.claude.com/docs/en/build-with-claude/refusals-and-fallback">new option</a> to request it falls back to another model automatically if something gets rejected.</p> <p>Claude Mythos 5 is out today as well, <a href="https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5-and-claude-mythos-5">Anthropic say it</a> "Shares Claude Fable 5's capabilities without the safety classifiers".</p> <p>The models have a 1 million token context window, 128,000 maximum output tokens and a knowledge cut-off date of January 2026.</p> <p>They are priced at twice the price of Claude Opus 4.5/4.6/4.7/4.8: $10/million input tokens and $50/million output tokens. There's no increase in price for longer context usage.</p> <p>Other than that the <a href="https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5-and-claude-mythos-5">upgrade guide</a> is substantially thinner than <a href="https://platform.claude.com/docs/en/about-claude/models/whats-new-claude-4-8">the similar guide for Opus 4.8</a>.</p> <h4 id="the-big-model-smell">The big model smell</h4> <p>The best way to describe Fable is that it feels <em>big</em>. Not just in terms of speed and cost, but also in how much it knows.</p> <p>Here's an example prompt, which I used to compare Fable's knowledge with Opus 4.8 (typo included):</p> <blockquote> <p><code>List all of Simon Willion's open source projects, most recent first, each with a rough date of when they were first released</code></p> </blockquote> <p>Opus 4.8's reply <a href="https://gist.github.com/simonw/e77abf645156ade8c5c610f524226dcb#response-1">started like this</a>:</p> <blockquote> <p>I don't have a reliable, comprehensive, and date-verified list of all of Simon Willison's open source projects, and I want to be honest about that rather than risk giving you inaccurate dates or fabricated entries.</p> <p>Here's what I can tell you with reasonable confidence:</p> <p><strong>Well-known projects by Simon Willison:</strong></p> <ul> <li> <p><strong>LLM</strong> (~2023) – A command-line tool and Python library for interacting with large language models.</p> </li> <li> <p><strong>Datasette</strong> (~2017) – A tool for exploring and publishing data, probably his most prominent recent project.</p> </li> <li> <p><strong>sqlite-utils</strong> (~2018-2019) – A Python library and CLI tool for manipulating SQLite databases, often used alongside Datasette.</p> </li> <li> <p><strong>Django</strong> (2005) – He is one of the co-creators of the Django web framework, along with Adrian Holovaty, originally developed at the Lawrence Journal-World newspaper.</p> </li> </ul> <p>He also maintains a large ecosystem of <strong>Datasette plugins</strong> and many smaller utilities (often with names like <code>shot-scraper</code>, various <code>*-to-sqlite</code> tools for importing data, etc.).</p> </blockquote> <p>That's not bad. Here's <a href="https://gist.github.com/simonw/e77abf645156ade8c5c610f524226dcb#response">the start of the response</a> for the exact same prompt from Fable 5 - both models were running without access to search:</p> <blockquote> <p>I think you mean <strong>Simon Willison</strong> — the prolific open source developer, co-creator of Django, and creator of Datasette.</p> <p>An important caveat upfront: a truly complete list isn't feasible. Simon Willison is famously prolific — he has published <strong>hundreds</strong> of repositories (Datasette alone has 100+ plugins, many written by him), and my knowledge has a cutoff date, so I'll be missing his most recent work. Below are his <strong>notable</strong> projects with approximate initial release dates, most recent first:</p> <h2 id="notable-projects-approximate-dates-">Notable projects (approximate dates)</h2> <ul> <li> <strong>files-to-prompt</strong> (April 2024) – concatenates files into a single prompt for LLMs</li> <li> <strong>datasette-extract</strong> (2024) – plugin for extracting structured data using LLMs</li> <li> <strong>LLM</strong> (May–June 2023) – CLI tool and Python library for interacting with large language models, plus a large ecosystem of plugins (llm-gpt4all, llm-mistral, llm-claude, etc., 2023 onward)</li> <li> <strong>symbex</strong> (June 2023) – search Python code for symbols</li> <li> <strong>ttok</strong> and <strong>strip-tags</strong> (May 2023) – CLI tools for token counting and HTML cleanup for LLM pipelines</li> <li> <strong>datasette-lite</strong> (May 2022) – Datasette running entirely in the browser via WebAssembly/Pyodide</li> <li> <strong>shot-scraper</strong> (March 2022) – automated website screenshots via Playwright</li> <li> <strong>s3-credentials</strong> (November 2021) – CLI for creating scoped AWS S3 credentials</li> <li> <strong>django-sql-dashboard</strong> (2021) – SQL reporting dashboards for Django</li> <li> <strong>Dogsheep suite</strong> (2019) – personal analytics tools: twitter-to-sqlite, github-to-sqlite, healthkit-to-sqlite, dogsheep-beta, etc.</li> <li> <strong>sqlite-utils</strong> (2018) – CLI and Python library for manipulating SQLite databases</li> <li> <strong>Datasette</strong> (November 2017) – his flagship project; tool for exploring and publishing data</li> <li> <strong>csvs-to-sqlite</strong> (2017) – convert CSV files to SQLite</li> <li> <strong>Various early tools</strong> (~2007–2010) – soupselect, json-head, geocoders, and others</li> <li> <strong>Django</strong> (developed 2003–2005, open-sourced July 2005) – co-created with Adrian Holovaty at the Lawrence Journal-World</li> </ul> </blockquote> <p>(Here's <a href="https://gist.github.com/simonw/8d05ce836d44bd8543055c9614f7c478">GPT-5.5 for good measure</a>. It listed even more projects than Fable did!)</p> <p>In the past I've stated that I don't care about how much models <em>know</em> - I want them to be able to manipulate text and code in useful ways and actively look up the information they need via search tools, not bake it into their weights.</p> <p>But knowledge like this is a reasonably good proxy for model size - you can cram a whole lot more details about the world into a larger number of parameters.</p> <p>Does <em>knowing more stuff</em> mean the model is better at the tasks we pose to it? I can certainly imagine how a coding model with deeper knowledge of modern libraries and patterns could crunch through coding tasks more effectively.</p> <p>Is Fable really bigger than Opus? Anthropic haven't said anything about model size, so all we have are tea-leaves, but the speed, pricing and my own poking at its knowledge make me think that it's a large model. Maybe the largest yet from any vendor.</p> <h4 id="using-fable-in-claude-ai">Using Fable in Claude.ai</h4> <p>Anthropic made Fable 5 available across all of their surfaces - the <a href="https://claude.ai/">Claude.ai</a> chat interface, Claude Code for web, Claude Code CLI and Claude Cowork as well. The model is available "until June 22nd" on the subscription plans (I'm on $100/month Max at the moment), after which it will be billed extra.</p> <p>Claude.ai is often under-estimated. Since <a href="https://simonwillison.net/2025/Sep/9/claude-code-interpreter/">September 2025</a> every chat has had access to a full container environment to run code, including the ability to install additional packages and even clone repositories directly from GitHub.</p> <p>Last week <a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/">I released micropython-wasm</a>, a Python library that uses <a href="https://wasmtime.dev/">wasmtime</a> to run a custom build of MicroPython in WebAssembly to act as a sandbox for untrusted Python code.</p> <p>I decided to see if Fable could upgrade that to running full Python instead. I started with this prompt:</p> <blockquote> <p><code>Clone simonw/micropython-wasm from GitHub and research how this could use a full Python as opposed to MicroPython</code></p> </blockquote> <p>Fable identified that it could use Brett Cannon's <a href="https://github.com/brettcannon/cpython-wasi-build/releases">cpython-wasi-build</a> builds for this, but was unable to download them itself due to environment restrictions.</p> <p>So I grabbed the two zip files from that page and uploaded them to Claude:</p> <blockquote> <p><code>Here's the Brett Cannon builds</code> (<code>python-3.zip</code>, <code>_build-python-3.zip</code> as attachments)</p> </blockquote> <p>And that was that. It churned away for a few minutes and got the entire thing working.</p> <p>Part of the response included:</p> <blockquote> <p>I tried the cleaner single-zip-stdlib approach to shrink the filesystem surface, but CPython's <code>getpath</code> bootstrap fails to find <code>encodings</code> from inside a zip without more prefix finessing — the directory-preopen approach works reliably, so that's what the PoC uses. The zip path is solvable but needs <code>_PYTHONHOME</code>/frozen-getpath work.</p> </blockquote> <p>So I said:</p> <blockquote> <p><code>Try a bit more at the single-zip-stdlib problem</code></p> </blockquote> <p>Then a little later:</p> <blockquote> <p><code>I want a wheel that has the whole system in it, the Python wrappers and the WASM files and the stdlibrary, so I can do uv run --with path-to-whl python -c "demo code"</code></p> </blockquote> <p>... and it gave me <a href="https://static.simonwillison.net/static/cors-allow/2026/cpython_wasm-0.1.0-py3-none-any.whl">this 13.9MB cpython_wasm-0.1.0-py3-none-any.whl</a> file. You can try running Python code in a sandbox using that wheel URL and <code>uv</code> like this:</p> <div class="highlight highlight-source-shell"><pre>uv run --with https://static.simonwillison.net/static/cors-allow/2026/cpython_wasm-0.1.0-py3-none-any.whl \ cpython-wasm -c <span class="pl-s"><span class="pl-pds">'</span>print(45 ** 56)<span class="pl-pds">'</span></span></pre></div> <p>Here's <a href="https://claude.ai/share/a73b8b8b-8ebc-4fef-9e5c-7438e5e7ae35">the full chat transcript</a>.</p> <p>This was a <em>very</em> strong start.</p> <h4 id="adding-features-to-datasette-agent-and-llm-using-claude-code">Adding features to Datasette Agent and LLM using Claude Code</h4> <p>Before I'd realized it was Fable day, my stretch goal for today was to add a new feature to <a href="https://agent.datasette.io/">Datasette Agent</a>: I wanted tool calls within that agent software to gain the ability to pause mid-execution and request approval directly from the user.</p> <p>This felt like a suitably meaty task to throw at the new model.</p> <p>Over the course of the day Fable not only <a href="https://github.com/datasette/datasette-agent/pull/20">solved that problem</a>, it also identified and then implemented four issues in my underlying LLM library that would help support this kind of advanced pause-resume mechanism in tool calls.</p> <p>It got everything working first using somewhat gnarly hacks, but the moment I told it that changes to LLM itself were in scope it set to work unraveling the hacks and turning them into supported features of LLM instead.</p> <p>My stretch goal turned into <a href="https://llm.datasette.io/en/latest/changelog.html#a3-2026-06-09">LLM 0.32a3</a>, almost entirely written by Fable. Here are the release notes:</p> <blockquote> <p>Driven by the needs of <a href="https://github.com/datasette/datasette-agent">Datasette Agent</a>'s human-in-the-loop <code>ask_user()</code> feature, made the following improvements to how tool calls work:</p> <ul> <li>Tool implementations can declare a parameter named <code>llm_tool_call</code> in order to be passed the <code>llm.ToolCall</code> object for the current invocation. This allows them to access the current <code>llm_tool_call.tool_call_id</code>. See <a href="https://llm.datasette.io/en/latest/python-api.html#python-api-tools-llm-tool-call">Accessing the tool call from inside a tool</a>. <a href="https://github.com/simonw/llm/pull/1480">#1480</a> </li> <li>Every tool call is now guaranteed a unique <code>tool_call_id</code> - providers that do not supply one get a synthesized <code>tc_</code>-prefixed ULID. <a href="https://github.com/simonw/llm/pull/1481">#1481</a> </li> <li>Tools can raise a <code>llm.PauseChain</code> exception to cleanly pause the tool chain, useful for things like waiting for human approval. The exception propagates to the caller with <code>.tool_call</code> and <code>.tool_results</code> (completed sibling results) attached, and no model call is made with a placeholder result. See <a href="https://llm.datasette.io/en/latest/python-api.html#python-api-tools-pause">Pausing a chain from inside a tool</a>. <a href="https://github.com/simonw/llm/pull/1482">#1482</a> </li> <li>Failure semantics for concurrent tool execution: async sibling tool calls always run to completion before a pause or hook exception propagates. <a href="https://github.com/simonw/llm/pull/1482">#1482</a> </li> <li>Chains can now resume from a <code>messages=</code> history ending in unresolved tool calls: the calls are executed through the normal <code>before_call</code>/<code>after_call</code> machinery before the first model call, skipping any that already have results. The <code>execute_tool_calls()</code> method also accepts a new optional <code>tool_calls_list=</code> argument for executing an explicit list of <code>ToolCall</code> objects in place of the calls requested by the response. See <a href="https://llm.datasette.io/en/latest/python-api.html#python-api-tools-resume">Resuming a chain with pending tool calls</a>. <a href="https://github.com/simonw/llm/pull/1482">#1482</a> </li> <li>Fixed a bug where the async tool executor silently dropped calls to tools not present in <code>tools=</code> - these now return <code>Error: tool "..." does not exist</code> results, matching the sync executor. <a href="https://github.com/simonw/llm/pull/1483">#1483</a> </li> </ul> </blockquote> <p>I'm really impressed with the quality of API design, tests, code and documentation that Fable put together for this. I spent several hours on it today, but it feels like several days' worth of work.</p> <h4 id="how-much-i-ve-spent">How much I've spent</h4> <p>I recently started using <a href="https://agentsview.io">AgentsView</a> to help track my local LLM usage across all of the different coding agents. I published a <a href="https://til.simonwillison.net/llms/agentsview-custom-model-price">TIL today</a> about adding custom Fable pricing to that tool, which I expect will not be necessary in the very near future.</p> <p>After setting the price, I ran this command to start a localhost web server to explore my usage:</p> <pre><code>uvx agentsview serve </code></pre> <p>Here's the treemap showing the breakdown of my Fable usage across various projects today:</p> <p><img src="https://static.simonwillison.net/static/2026/agentsview-fable-full-day.jpg" alt="Screenshot of a cost tracking dashboard with two panels. The first panel is titled &quot;Cost Attribution&quot; with toggle buttons for Project / Model / Agent and Treemap / List, with Project and Treemap selected. Italic text reads &quot;Click to hide from chart&quot;. A treemap shows a large red block labeled prod_datasette_agent $99.26 89.9%, with smaller blocks to its right labeled cloud (blue), datasette (teal), llm (red), and money (pink), plus a tiny orange sliver. A legend lists: 1 prod_datasette_agent $99.26, 2 cloud $3.98, 3 datasette $2.81, 4 llm $2.30, 5 money $1.92, 6 simon $0.15. The second panel is titled &quot;Top Sessions by Cost&quot; and lists nine sessions, each with a &quot;Claude&quot; badge, a prompt excerpt, a project name with a session UUID (omitted here), a token count, and a cost: 1. Review ./datasette-agent and ./datasette-apps - we are going to add a new feature to agent but you ... prod_datasette_agent, 78.2M, $99.26. 2. issues.db is a copy of the Datasette issues database. There are a LOT of notes in there relating to... datasette, 826.8k, $2.81. 3. Consult fly-docs and then look at datasette.cloud (which launches fly machines) and datasettecloud-... cloud, 924.7k, $2.61. 4. simonwillisonblog.db is a copy of my blog, plus all my software releases and other interesting thin... money, 542.9k, $1.92. 5. Look in datasette.cloud and figure out all remaining steps and decisions that need to be made in or... cloud, 455k, $1.37. 6. Review PRs and issues filed against this repo within the last 4 weeks and see if any deserve to be ... llm, 323.3k, $0.95. 7. run mypy, llm, 320.9k, $0.76. 8. [Image #1] fix this in github actions, llm, 183.9k, $0.59. 9. simon, simon, 26.4k, $0.15." style="max-width: 100%;" /></p> <p>I used $110.42 worth of tokens today, all as part of my $100/month subscription.</p> <h4 id="and-some-pelicans">And some pelicans</h4> <p>I ran "Generate an SVG of a pelican riding a bicycle" against all five thinking effort levels with Fable.</p> <p>Here are <a href="https://tools.simonwillison.net/markdown-svg-renderer#url=https%3A%2F%2Fgist.github.com%2Fsimonw%2F94fde31c34a0400c1d29f57e6a708e6b">the results</a>, including the token cost for each one:</p> <div style="display: flex; flex-wrap: wrap; gap: 10px; margin-bottom: 1em"> <figure style="margin: 0; flex: 1 1 30%;"> <img src="https://static.simonwillison.net/static/2026/fable-low.jpg" alt="low" style="width: 100%; height: auto;" /> <figcaption style="text-align: center;">low: <a href="https://www.llm-prices.com/#it=25&amp;ot=1929&amp;sel=claude-fable-5">1,929 out, 9.67c</a></figcaption> </figure> <figure style="margin: 0; flex: 1 1 30%;"> <img src="https://static.simonwillison.net/static/2026/fable-medium.jpg" alt="medium" style="width: 100%; height: auto;" /> <figcaption style="text-align: center;">medium: <a href="https://www.llm-prices.com/#it=25&amp;ot=2290&amp;sel=claude-fable-5">2,290 out, 11.475c</a></figcaption> </figure> <figure style="margin: 0; flex: 1 1 30%;"> <img src="https://static.simonwillison.net/static/2026/fable-high.jpg" alt="high" style="width: 100%; height: auto;" /> <figcaption style="text-align: center;">high: <a href="https://www.llm-prices.com/#it=25&amp;ot=2057&amp;sel=claude-fable-5">2,057 out, 10.31c</a></figcaption> </figure> <figure style="margin: 0; flex: 1 1 45%;"> <img src="https://static.simonwillison.net/static/2026/fable-xhigh.jpg" alt="xhigh" style="width: 100%; height: auto;" /> <figcaption style="text-align: center;">xhigh: <a href="https://www.llm-prices.com/#it=25&amp;ot=5992&amp;sel=claude-fable-5">5,992 out, 29.985c</a></figcaption> </figure> <figure style="margin: 0; flex: 1 1 45%;"> <img src="https://static.simonwillison.net/static/2026/fable-max.jpg" alt="max" style="width: 100%; height: auto;" /> <figcaption style="text-align: center;">max: <a href="https://www.llm-prices.com/#it=25&amp;ot=14430&amp;sel=claude-fable-5">14,430 out, 72.175c</a></figcaption> </figure> </div> <p>It's interesting that high ended up using fewer tokens than medium for this particular run.</p> <p>Here are the <a href="https://simonwillison.net/2026/May/28/claude-opus-4-8/#and-some-pelicans">Opus 4.8 pelicans</a> for comparison.</p> <p>Tags: <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/anthropic">anthropic</a>, <a href="https://simonwillison.net/tags/claude">claude</a>, <a href="https://simonwillison.net/tags/llm-pricing">llm-pricing</a>, <a href="https://simonwillison.net/tags/pelican-riding-a-bicycle">pelican-riding-a-bicycle</a>, <a href="https://simonwillison.net/tags/llm-release">llm-release</a>, <a href="https://simonwillison.net/tags/claude-mythos">claude-mythos</a></p>

A new FDA-approved ingredient can make US sunscreens more efficient and luxurious-feeling. LordHenriVoton/Getty Images The FDA just approved bemotrizinol (BEMT), a popular ingredient in European and Asian sunscreens. This is the first time the FDA approved a new sunscreen ingredient since 1999. For decades, US sunscreen lagged behind other countries' in effectiveness and smoothness. You no longer have to haul French sunscreens home from your vacation — or pay exorbitant shipping fees — to get your hands on the good stuff. The FDA just approved bemotrizinol (BEMT), an ingredient used in European and Asian sunscreens for over 20 years. It's the first upgrade to US sunscreen this century. BEMT is a chemical UV filter that absorbs both UVA and UVB rays, and doesn't break down easily when exposed to sunlight, making it more effective at protecting against skin cancer and preventing wrinkles. Due to regulatory differences, the US's sunscreen formulas have remained notoriously old-fashioned, lagging behind other countries like the UK, South Korea, Japan, and Australia, which have been proactive in approving new, advanced UV-blocking ingredients. As a result, sunscreens sold in these countries offer better protection against both UVA and UVB rays while also providing smoother application. That's about to change — but consumers will have to wait over a year to get their hands on these new and improved American products. The biggest FDA filter update since 1999 It's taken decades to approve new sunscreen ingredients in the US. Drazen Zigic/Getty Images The FDA treats sunscreens as non-prescription drugs — meaning each ingredient has to go through a lengthy regulatory approval process. While this sounds good in theory, it's led to slow-moving sunscreen advancement; the FDA hasn't approved a new UV-blocking filter in decades. It's not to say that US sunscreens are completely ineffective. According to Dr. Ellen Gendler, a dermatologist practicing in New York City, US sunscreens are "very good" at blocking UVB rays — the ones that cause sunburn. However, she said we only have only one ingredient approved for UVA blockage: Avobenzone, which she described as "not very stable" and in need of other ingredients to make it effective. UVA rays are deeper-penetrating, which is why you can still get a tan while wearing US sunscreen — something Gendler noted might be "even more harmful to the skin" than the damage done by UVB rays, because UVA rays are known to cause skin cancer. SPF also isn't an indicator of protection against UVA. "I have patients use Neutrogena all the time, and they say, 'I'm using an SPF 70', and they're still tan," she said. "That's because they're not well protected against ultraviolet A." The search for a sleeker formula A common complaint of US sunscreens is that they leave a white cast. Meeko Media/Getty Images Unlike in the US, sunscreens in other countries are classified as cosmetics and not non-prescription drugs, making them subject to more lax regulations and faster innovation. Beyond being more effective at protecting the skin, they also cause less of a white cast and feel less greasy. Gendler advises her patients to layer chemical and mineral sunscreens with zinc oxide for better protection — but even that "doesn't afford them the same broad spectrum as the European products do." For example, her favorite international sunscreen is La Roche Posay's SPF 50 Anthelios UVMune 400 sunscreen, which comes in multiple formulas such as "invisible" and "tinted." It uses Mexoryl 400, an advanced version of Mexoryl, which has been considered one of the best UVA-blocking ingredients since 2005. However, the FDA has only approved some forms of Mexoryl in limited formulas. That means La Roche Posay's US versions of the same sunscreen don't include Mexoryl. In the US, the only La Roche Posay product with Mexoryl is an SPF-15 moisturizer. Bemotrizinol can make for stronger, smoother sunscreens US sunscreens may start to resemble those in Europe and Asia. mihailomilovanovic/Getty Images Besides offering more UV protection than some other sunscreen ingredients, bemotrizinol is also supposed to allow for a more seamless, streak-free application. Still, it'll take a while for new sunscreen formulas to include the ingredient. DSM Nutritional Products, which filed for FDA approval of bemotrizinol as a sunscreen ingredient in 2024, will have exclusive rights to sell their proprietary BEMT formulation in the US for 18 months, according to Vox. After that, other cosmetic companies will likely follow suit. In the meantime, if you have a vacation abroad coming up, you can always leave some extra room in your luggage on your way back home. Or, you can always call in a favor. "If your friends are going to Europe, instead of bringing you back a scarf, tell them to bring you back sunscreen," Gendler said. Read the original article on Business Insider

Ella Feldman / Bloomberg: CoreWeave founders have sold $2.3B in stock since the company's lockup period expired in August after a March 2025 IPO, reducing their combined holdings by ~25%  —  CoreWeave Inc.'s stock price has more than doubled since the AI data center operator's March 2025 initial public offering.

Instead of spending a year raising a formal venture fund, the Sabertooth VC founder used a captive network of LPs to invest in startups like Anthropic, Anduril, and SpaceX.

Lawrence Abrams / BleepingComputer: ServiceNow says attackers exploited a flaw, patched on June 5, that let unauthenticated users query data from customer instances, but gives few other details  —  ServiceNow is warning about a security incident after attackers exploited an unauthenticated access flaw through a vulnerable API endpoint …

Akash Sriram / Reuters: Sources: SpaceX aims to launch orbital AI computing tests by late 2027 and has requested permission from regulators to launch up to 1M data-center satellites  —  SpaceX executives say the company is aiming to launch initial demonstrations of space-based artificial intelligence computing infrastructure …

While Fable 5 excels in long-term tasks, its complex processes can lead to slower performance. The AI market’s increased focus on reasoning is driving up the cost of these models.

OpenAI CEO Sam Altman is at a pivotal moment as the company marches toward an IPO. Eric Lee/Bloomberg via Getty Images OpenAI announced on Monday that it had taken the first official step toward an IPO. OpenAI's entry now means that three major AI companies are in the process of going public, potentially the same year. Financial analysts portrayed the IPO rush as yet another front in the AI race. The next phase of the AI race is on. With OpenAI's Monday confidential S-1 filing following in the footsteps of Antrhopic's, the two premiere AI labs are now officially in the IPO race. SpaceX, which includes Elon Musk's xAI, has a head start, and goes public later this week. While OpenAI left the door open on its timing, warning that "it may be a while" before it actually publicly lists its shares, the AI labs are moving from an era of closed-door tinkering as a private companies into the world of public earnings calls, financial disclosures, and the regulation that comes with being a publicly traded company. Is it a telling signal of just how much cash they're all burning? A chance to reward loyal employees who hold shares and help retain key AI talent? Something else? Here's what smart voices across Wall Street, tech, and the wider business world make of the ChatGPT maker's latest move: Aravind Srinivas, CEO of Perplexity Perplexity CEO Aravind Srinivas Bloomberg/Getty Images Perplexity CEO Aravind Srinivas said his AI company is still targeting 2028 for its IPO. "Agnostic of these two companies, we were planning for something in 2028, so that still remains the case," Srinivas told CNBC. Srinivas said how OpenAI, Anthropic, and SpaceX's IPOs fare will matter to the broader industry. "I certainly think there will be ripple effects if they don't go well, like there is no sugar coating on that," he said. Dan Ives, tech analyst at Wedbush Securities Dan Ives, Wedbush's managing director and senior equity research analyst covering the tech sector. NASDAQ: ORBS) Wedbush analyst Dan Ives said OpenAI's filing shows "the floodgates for the IPO market are officially open." "While both OpenAI and Anthropic have recently raised significant amounts of capital, both are racing to get to market as quickly as possible to beat one another out due to the significant amount of capital both firms are looking to raise," Ives wrote in a note. Ives said OpenAI CEO Sam Altman "continues to face competitive pressures from competitors like Anthropic to show that growth continues to accelerate and will maintain its current aggressive pace through the next decade." Dan Niles, founder and portfolio manager at Niles Investment Management Dan Niles said he views Anthropic more favorably than OpenAI. "I think Google wins in consumer, they have the complete stack," the portfolio manager told CNBC. "I think they win in AI overall, but then in corporate, you have Anthropic." Niles said that Anthropic reached profitability in Q2 "and their revenues are ramping like nothing you've ever seen in history for a company of that size." "I think OpenAI is stuck between the two," he said. "So for me, those are the two winners." Michael Fertik, founder of Verdict Capital Michael Fertik, founder of Verdict Capital Chris J. Ratcliffe/Bloomberg via Getty Images Michael Fertik, founder of Verdict Capital, said he hopes SpaceX's IPO on Friday opens the door "to a gushing torrent of liquidity." "I'm rooting for Elon Musk. I'm rooting for the IPOs. I'm rooting for the OpenAIs and Anthropics," Fertik told CNBC. Fertik said that companies debuting strongly would help the US stay ahead in the generative AI race. He also said that many companies have been relying on private markets for so long that it's time for retail investors to share in the AI moment. "I hope that it does encourage companies that have been getting ready to file, to file," he said of SpaceX's debut. Gregory Allen, founder and CEO of Decision Tree Research Gregory Allen, CEO of Decision Tree Research, said OpenAI's valuations, along with those of SpaceX and Anthropic, are really extraordinary. "These companies are in the ballpark of a trillion dollars valuation," Allen told CNBC, adding that in annuity terms, that is akin to "an annuity that kicks out $45 billion a year every year forever." Allen said even though companies like OpenAI are going to lose money this year, "the gross margins in these businesses are extremely favorable." "The question is, can they time the investments correctly, to make these, hundreds of billions of dollars in capital expenditure investments and ride the wave of revenue growth properly, such that they don't run out of cash and go bankrupt, which is the big risk here," he said. Gary Marcus, AI researcher and professor emeritus at New York University AI researcher Gary Marcus Matthias Balk/picture alliance via Getty Images Gary Marcus, an AI researcher and author known for his critical views of the AI industry, suggested buyers should be aware of where the AI industry stands. "i used to be the only person saying the emperor was wearing no clothes," Marcus wrote on X, quoting a tweet about SpaceX's valuation. "those days are over. the only people buying SpaceX and OpenAI at list price will be the stragglers who still can't see it." Nate Elliott, principal analyst at EMARKETER Nate Elliot of EMARKETER Courtesy of EMARKETER OpenAI "is filing to go public at a precarious moment," said Nate Elliott, a principal analyst at EMARKETER, a sister company to Business Insider. "The company is just about to lose its strong early leads in both consumer and enterprise AI," Elliott wrote in an email to BI. "Anthropic has generated incredible momentum in the enterprise market, and EMARKETER forecasts that Google will overtake ChatGPT in terms of US AI users by early 2027." Elliott said OpenAI "doesn't have a lot of other places to look for the enormous capital required to support its costs." "The company has taken almost $200 billion in funding across more than a dozen private rounds," he wrote. "It certainly can't count on its consumer business becoming self-sufficient anytime soon." Joshua Achiam, OpenAI's chief futurist It wouldn't be a full OpenAI news cycle without someone taking a shot at Anthropic. Joshua Achiam, OpenAI's chief futurist, framed the discussion as a choice between two very different goals of the now rival companies. "Should a loving ensouled machine God watch over humanity? Vote Anthropic," he wrote on X. "Should humanity be entrusted with the tools of its own progress and destiny? Vote OpenAI." Achiam dismissed the analysis that focuses on the differences between the two leading AI companies' business models. "If your lens for analyzing this is 'consumer v enterprise business' your ability to understand what's going on is unfixably borked," he wrote. Read the original article on Business Insider

<p><strong>Release:</strong> <a href="https://github.com/simonw/llm/releases/tag/0.32a3">llm 0.32a3</a></p> <p>Almost entirely written by the new Claude Fable 5, see <a href="https://simonwillison.net/2026/Jun/9/claude-fable-5/#adding-features-to-datasette-agent-and-llm-using-claude-code">my write-up for more details</a>.</p> <p>Tags: <a href="https://simonwillison.net/tags/projects">projects</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/llm">llm</a>, <a href="https://simonwillison.net/tags/claude-mythos">claude-mythos</a></p>

Former National Cyber Director Chris Inglis warns that cyberattacks threaten hospitals, utilities, and essential services.

A US Apache operating in the Middle East. US Marine Corps photo Iran shot down a US Army Apache helicopter in the Middle East, President Donald Trump said on Tuesday. The US military used a new naval drone, a Saronic Corsair, to rescue the two crew members, a defense official said. The drone's participation appears to be an operational first for the US military. The US military used a new naval drone to help rescue two American soldiers after Iran shot down their Apache helicopter off the coast of Oman. US Central Command, which oversees military operations in the Middle East, said on Tuesday that American forces had rescued two US Army AH-64 Apache crew members after their helicopter went down while patrolling regional waters. The soldiers were rescued within roughly two hours and are in stable condition, CENTCOM said in a statement, adding that the cause of the aircraft loss was under investigation. Later on Tuesday, President Donald Trump said on Truth Social that Iran "shot down" the Apache while it was patrolling over the Strait of Hormuz, vowing the US would "respond to this attack." US forces began launching "self-defense strikes" against Iran several hours later, the military said, calling it "a proportional response." Navy Capt. Tim Hawkins, a CENTCOM spokesperson, identified the drone that assisted in the Apache rescue mission as a Corsair uncrewed surface vessel operated by Task Force 59, a Middle East-based unit focused on integrating cutting-edge technologies into maritime operations. The incident this week marks the US military's first Apache loss of the Iran war. US Army Reserve photo by Pfc. Luis Daniel Torres Hawkins said the naval drone picked up the soldiers and moved them to another location on the water, where they were then lifted into a helicopter for further transport. He said that "the task force began fielding these drones in theater in late March." The drone's participation in the rescue mission appears to be an operational first for the military and comes amid a broader Pentagon push to integrate uncrewed systems into operations. In its earlier statement, CENTCOM said that rescue efforts were led, in part, by the Army's 82nd Airborne Division, an elite conventional infantry division that specializes in parachute-assault operations and rapid crisis response. Thousands of soldiers from the 82nd deployed to the Middle East earlier in Operation Epic Fury, which began at the end of February, as the Trump administration weighed sending troops to conduct ground operations in Iran. CENTCOM said that the Apache crew rescue mission was also supported by the US Air Force and US Navy. Business Insider previously visited Task Force 59 at the Navy's 5th Fleet headquarters in Bahrain, observing some of the uncrewed surface vessels available to the task force. Jake Epstein/Business Insider The involvement of Corsair — a 24-foot-long surface drone developed by Texas-based Saronic Technologies — in Tuesday's rescue mission is a significant development for Task Force 59, which was established in September 2021 and has been experimenting with uncrewed capabilities. The US military has lost dozens of aircraft — including fighter jets, refueling tankers, helicopters, and combat drones — in the Middle East since the start of the Iran war; however, the latest incident marks the first Apache loss in the region since the conflict began. These well-known attack helicopters, flown by the American military and several US allies, can be armed with AGM-114 Hellfire missiles and have been used to target Iranian small boats around the Strait of Hormuz. The loss comes amid a fragile, monthslong ceasefire between the US and Iran that has been stressed by exchanges of fire. The two sides have repeatedly traded fire in recent days, despite Trump's insistence that a deal is near. On Friday, for instance, the US military said it carried out strikes on Iranian radar sites after intercepting Iranian ballistic missiles and drones launched toward the Strait of Hormuz and Gulf states. And on Monday, Israel and Iran traded strikes, further testing the ceasefire. Read the original article on Business Insider

Voluminous patch updates could soon be the norm, as artificial intelligence accelerates the speed and scale of vulnerability discovery.

GM said it has 250,000 EVs on US roads that could be turned into an energy source for power grids. Courtesy General Motors The AI data center buildout is driving a historic surge in electricity demand. General Motors proposes leveraging personal EVs to help balance the power grid. A GM executive told Business Insider this program could result in lower utility bills. General Motors wants to use the parked EV in your garage to power the energy grid and potentially lower your utility bills. The legacy automaker is pitching EV batteries as a power source that could help utility companies manage surging electricity demand, as households see rate increases and the US sees historic levels of energy consumption amid the AI data center build-out. In an open letter to utility executives and energy lawmakers on Tuesday, GM called on utility agencies, regulators, and automakers for a "public-private collaboration" that could make it easier for EV drivers to send power back to the grid. Stakeholders would work out logistics like customer enrollment, incentives, and rate structures, the company said. "The growth and consumption of electricity in the US has been at historic highs," Kurt Kelty, GM's vice president of battery and sustainability, told Business Insider. "We've never had such a high growth rate in electricity for probably a hundred years or so. That's partly being driven by the AI data center growth." The US Energy Information Administration said in January that it expects electricity use to rise 1% this year and 3% in 2027, marking the strongest four-year stretch of demand growth since 2000. The agency said "demand from large computing centers" was behind the surge. A Business Insider analysis of US data center permits found that if all facilities permitted through 2025 came online, they could consume between 224.3 terawatt-hours and 358.8 terawatt-hours of electricity a year. At the midpoint, that's more than the total electricity use by any US state outside Texas in 2024. To ease grid demand, GM said it could leverage bidirectional charging technology, which allows power to flow into and out of the EV. The company said it has more than 250,000 bidirectional EVs on US roads and is committed to building that capability across its entire EV portfolio. With a quarter of a million EVs, GM said that amounts to power for about 120,000 homes for up to one week or all of San Francisco for one day. GM's energy home system Courtesy of GM Kelty described the future customer experience as a kind of "buy low, sell high" scenario for electricity. A driver could come home from work, plug in, and let the EV discharge power during peak demand times, such as late afternoon or evening. Then the car could recharge later at night, when electricity is cheaper. Kelty added that this process wouldn't happen every day and that utility companies may only need the extra support 20 or 30 days out of the year. The EV would also not be fully discharged each time. "It's not that much extra wear and tear on the battery," he said, adding that the battery would still be covered under warranty as long as customers followed a GM-specific program. Whether customers will see returns through credits, lower utility bills, or a cheaper EV at initial purchase is all on the table, Kelty said. The idea is still in its early phases. GM said it's now testing "vehicle-to-grid" programs with utility agencies, including PG&E in northern California and DTE Energy in Michigan. Tesla has a similar concept with Powershare, which allows Cybertruck owners to use the truck's battery to power a home and, in some regions, support the grid. However, the technology is currently only centered on the Cybertruck. "The big thing we're looking at is we want to sell more EVs," Kelty said. "We want to reduce fossil fuel consumption. That's the direction we want to go." Read the original article on Business Insider

<p><strong>TIL:</strong> <a href="https://til.simonwillison.net/llms/agentsview-custom-model-price">Setting a custom price for a model in AgentsView</a></p> <p>I've been really enjoying <a href="https://agentsview.io/">AgentsView</a> by Wes McKinney as a tool for exploring my token usage across different coding agents running on my laptop.</p> <p>Claude Fable 5 came out today and wasn't yet included in the pricing database AgentsView uses. I used Fable to reverse-engineer AgentsView and figured out this recipe for setting custom prices.</p> <p>Here's my Claude Fable 5 usage for today so far, plotted by AgentsView as a treemap across my different local projects:</p> <p><img alt="Screenshot of a cost analytics dashboard. Cost Attribution - Click to hide from chart - toggle buttons for Project / Model / Agent and Treemap / List. A treemap shows a large red block: prod_datasette_agent $74.06 89.3%, then blue: cloud $3.98 4.8%, teal: datasette $2.81 3.4%, pink: money $1.92 2.3%, and a thin orange sliver. A legend lists 1 prod_datasette_agent $74.06, 2 cloud $3.98, 3 datasette $2.81, 4 money $1.92, 5 simon $0.15. Below left, Top Sessions by Cost: 1 Claude - Review ./datasette-agent and ./datasette-apps - we are going to a... - prod_datasette_agent · 08a1f374-0e77-420f-be2d-af805d67e8aa - 55.9M $74.06; 2 Claude - issues.db is a copy of the Datasette issues database. There are a... - datasette · 8caa2d2d-b91f-43b3-bf3a-4268995b6011 - 826.8k $2.81; 3 Claude - Consult fly-docs and then look at datasette.cloud (which launche... - cloud · bfcacc70-09d7-4b27-aaec-4bb8accd9fec - 924.7k $2.61; 4 Claude - simonwillisonblog.db is a copy of my blog, plus all my software re... - money · 0c0fb9dc-6347-4e1b-9307-3709a7cdf0c8 - 542.9k $1.92; 5 Claude - Look in datasette.cloud and figure out all remaining steps and dec... - cloud · 45963b5f-608a-4caa-ad6b-6ae81e1dbf0d - 455k $1.37; 6 Claude - simon - simon · deeccb5d-9e90-4b1e-bfe6-c2b271e1b1d4 - 26.4k $0.15. Below right, Cache Efficiency with horizontal bars: Cache Reads 57.6M (nearly full green bar), Cache Writes 769.3K, Uncached Input 64.4K, Output 300.9K (all tiny bars), and a green highlighted note: $516.62 saved vs uncached." src="https://static.simonwillison.net/static/2026/agentsview-fable.jpg" /></p> <p>Tags: <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/llm-pricing">llm-pricing</a>, <a href="https://simonwillison.net/tags/claude-mythos">claude-mythos</a></p>

Salesforce Benjamin Fanjoy/Getty Images Salesforce laid off employees in a new round of cuts this week. An internal document shows Salesforce's standard severance offer for U.S. employees. Employees receive severance based on level, tenure, and age. As Salesforce begins another round of layoffs, the company's standard severance plan appears to offer employees a softer landing than many of its Big Tech peers. Internal documents reviewed by Business Insider show that eligible U.S. workers can receive up to 30 weeks of severance, a package that appears more generous than those recently offered by companies including Oracle, Amazon, and Block. Salesforce on Monday started notifying employees about layoffs, according to people familiar with the matter and a regulatory notice in California. The layoffs come amid concerns that AI tools and agents could replace traditional software, including the company's main customer relationship-management offering. Salesforce's stock is down more than 30% this year. Salesforce's severance policy states that employees receive severance payments based on level, tenure, and age. Senior directors and director-level employees receive 13 weeks of base pay, while senior managers and below receive 9 weeks. Employees ages 60 and older at any level get an additional four weeks. Employees also receive an additional three weeks of service per year. A partial year counts as a full year. Employees receive six months of COBRA coverage, or 12 months if they are 60 or older. The combined total of level- and tenure-based is capped at 26 weeks, or 30 weeks for those 60 and older. Salesforce's package appears larger than other recent Big Tech severance offers. Oracle recently offered laid-off US employees four weeks' base salary, plus one week per additional year of employment, up to 26 weeks, as severance. Block, which laid off nearly half its employees earlier this year, provided them with 20 weeks of salary, plus one additional week per year of tenure. Amazon's package for laid-off employees in January included full pay and benefits for 90 days, plus an additional severance package. Salesforce did not respond to requests for comment. Have a tip? Contact this reporter via email at astewart@businessinsider.com or Signal at +1-425-344-8242. Use a personal email address and a nonwork device; here's our guide to sharing information securely. Read the original article on Business Insider

Starlink, SpaceX's top moneymaker, also raised service prices by $5 to $10.

GM is developing an entirely new sodium-ion battery chemistry for use in everything from data centers to its own factories.

A separate zero-day also disclosed by Nightmare Eclipse appears to be patched as well.

Discussion | Link

I'm desperate for a personal AI assistant, but do I really want to become the kind of person who can't function without the friendly robot voice in my phone?