Releases·Tool Release
Datasette 1.0a39 and 0.65.4 security releases fix bugs found by Claude Fable 5.1, GPT-5.6, and GPT-6 Astra audit
Simon Willison released Datasette 1.0a39 and 0.65.4 patching security flaws found by auditing the codebase with frontier LLMs (Claude Fable 5.1, GPT-5.6, GPT-6 Astra), after issues reported by Sevban Dönmez. He and Alex Garcia reviewed fixes for nearly a week and plan to make frontier-model security audits standard practice.
0 engagement1 sourceFri, Sep 11, 2026, 03:27 AM
AI Overview
- Simon Willison and Alex Garcia released security fixes for Datasette instances mixing public and private tables, responding to issues reported by Sevban Dönmez
- The pair audited Datasette using Claude Fable 5.1, GPT-5.6, and GPT-6 Astra, then spent nearly a week collaborating on and reviewing fixes
- They split work so one person wrote automated tests exposing each issue while the other implemented the fix, with two humans reviewing each issue
- Willison says frontier-model security audits will be incorporated into all future Datasette development
Related
Social·Communityfeatured
Matt Wolfe posts first hands-on look at GPT-6 Astra; Reddit users report strong early impressions at global rollout
OpenAI's GPT-6 Astra rolled out globally around September 4-5, 2026, drawing a wave of hands-on reactions across Reddit, YouTube, and blogs. Users report fewer clarification prompts, better visual essays than GPT-5.6 Sol, and 100 performance improvements found in a 45k+ LOC Rust codebase. A recurring theme is token efficiency: Artificial Analysis data cited in one post shows Astra using roughly 21k output tokens per task versus about 64k for Claude Fable 5.1.
OpenAI's GPT-6 Astra rolled out globally around September 4-5, 2026, drawing a wave of hands-on reactions across Reddit, YouTube, and blogs. Users report fewer clarification prompts, better visual essays than GPT-5.6 Sol, and 100 performance improvements found in a 45k+ LOC Rust codebase. A recurring theme is token efficiency: Artificial Analysis data cited in one post shows Astra using roughly 21k output tokens per task versus about 64k for Claude Fable 5.1.
Updated Wed, Sep 9, 2026, 03:16 AM — Artificial Analysis token-usage data surfaced: Astra ~21k output tokens per task vs ~64k for Claude Fable 5.1.
83212.0k engagement16 sourcesyoutube, reddit, rssFri, Sep 4, 2026, 12:00 AM
News·Jobs
Datasette creator shows coding agent impact via GitHub code-frequency chart
Simon Willison published a GitHub code-frequency chart for his Datasette project, showing a dramatic spike in code activity coinciding with the release of advanced AI coding models like Opus 4.8, GPT-5.5, Fable 5, and GPT-5.6 Sol. The chart illustrates the tangible impact of AI coding agents on developer output.
Simon Willison published a GitHub code-frequency chart for his Datasette project, showing a dramatic spike in code activity coinciding with the release of advanced AI coding models like Opus 4.8, GPT-5.5, Fable 5, and GPT-5.6 Sol. The chart illustrates the tangible impact of AI coding agents on developer output.
0 engagement1 sourcerssMon, Jul 13, 2026, 09:45 PM
Social·Community
GPT-6 Astra vs Claude Fable 5.1: early benchmarks show mixed results at matching $10/$50 per M token pricing
Following OpenAI's September 4 release of GPT-6 Astra and Anthropic's Claude Fable 5.1 three days earlier, users compiled benchmark comparisons showing mixed results. Newer analyses argue both labs' launch tables are accurate but use nearly non-overlapping benchmark suites, so each makes its model look dominant. Real cost also diverges despite identical headline pricing, with Astra using fewer tokens per task.
Following OpenAI's September 4 release of GPT-6 Astra and Anthropic's Claude Fable 5.1 three days earlier, users compiled benchmark comparisons showing mixed results. Newer analyses argue both labs' launch tables are accurate but use nearly non-overlapping benchmark suites, so each makes its model look dominant. Real cost also diverges despite identical headline pricing, with Astra using fewer tokens per task.
Updated Thu, Sep 10, 2026, 02:54 PM — Users conclude both labs' launch benchmark tables are accurate but use nearly non-overlapping suites, so neither head-to
051138 engagement6 sourcesredditFri, Sep 4, 2026, 03:26 AM
Products·launched
Browser-based video compressor using WebAssembly FFMPEG
A web tool at tools.simonwillison.net that compresses videos entirely in the browser. Simon Willison had Claude Fable 5.1 in Claude Code build it using the WebAssembly build of FFMPEG, to optimize phone-recorded demo videos for his blog.
A web tool at tools.simonwillison.net that compresses videos entirely in the browser. Simon Willison had Claude Fable 5.1 in Claude Code build it using the WebAssembly build of FFMPEG, to optimize phone-recorded demo videos for his blog.
0 engagement1 sourcerssMon, Sep 7, 2026, 06:29 PM
Social·Community
Generating running routes with GPT-6 Astra and ChatGPT Work
Simon Willison used ChatGPT Work with GPT-6 Astra (Max) to generate looping 5K and 10K running routes from his address using OpenStreetMap data. The agent ran for 27 minutes and produced embedded map visualizations plus downloadable GPX and GeoJSON files. He criticizes the ChatGPT UI for hiding the code the agent ran and for compaction destroying access to it.
Simon Willison used ChatGPT Work with GPT-6 Astra (Max) to generate looping 5K and 10K running routes from his address using OpenStreetMap data. The agent ran for 27 minutes and produced embedded map visualizations plus downloadable GPX and GeoJSON files. He criticizes the ChatGPT UI for hiding the code the agent ran and for compaction destroying access to it.
0 engagement1 sourcerssSat, Sep 12, 2026, 11:56 PM