llm 0.37.3
LLM 0.37.3 is a bug-fix release addressing a break caused by Homebrew shipping SQLite 3.54.1, whose correct handling of scalar subqueries in json_object() turned nested arrays into strings and broke LLM's logging code.
Reporting and perspectives from Simon Willison. Headlines and excerpts link to the original articles.
LLM 0.37.3 is a bug-fix release addressing a break caused by Homebrew shipping SQLite 3.54.1, whose correct handling of scalar subqueries in json_object() turned nested arrays into strings and broke LLM's logging code.
Simon Willison released datasette-maplibre 0.1a0, a Datasette geospatial plugin that renders GeoJSON shapes and lat/lon points on maps using MapLibre GL with OpenFreeMap's free vector tiles, mostly built via a single prompt to Claude.
datasette-apps 0.4a0 adds URL fragment linking, letting sandboxed HTML/JavaScript apps hosted inside Datasette read #fragments and discover supported URL schemes so apps can link to specific records in other apps.
datasette-llm 0.1a11 adds support for OpenAI-compatible endpoints, letting Datasette point at local models served by LM Studio, Ollama or llama.cpp via an api_base setting.
datasette-vega 0.7 revives Datasette's chart plugin after its first update since July 2018, adding compatibility with the latest Datasette alphas and cosmetic fixes for Datasette Lite, which can now run JavaScript plugins like this one.
datasette-leaflet 0.3 fixes Datasette pages that broke OpenStreetMap's referrer policy by switching tile requests to strict-origin-when-cross-origin. The policy now sends the host but not the full page path.
Simon Willison released llm-gemini 0.37, adding support for Google's new Gemini 4 Hydrogen, the cheapest Gemini 4 model at $1.50/$7.50 per million input/output tokens, though only 25% below Carbon and Argon until their prices double on December 1, 2026.
Google has released Gemini Decision 1, a decision-model API that answers yes/no, multiple-choice and rubric-scoring questions and unlike Jev also accepts images, audio, video and PDFs. It is priced at 10 cents per million input tokens with free output, matching OpenAI's GPT-6 Luna.
AWS has released an official aws-emulator that lets developers test against a local clone of its services without provisioning a paid API key; it is written in Rust and accumulated 29,684 commits in its first three weeks.
Simon Willison added tag-cloud browsing to tools.simonwillison.net, using tree-sitter to extract JavaScript APIs, libraries and web API hosts from each of his 244 (and counting) vibe-coded tools.
llm 0.37.2 fixes two paths where an OpenAI API key could be sent to a third-party OpenAI-compatible endpoint configured through extra-openai-models.yaml, reported by Daniel de Araujo in two GitHub security advisories.
Anthropic's IPO roadshow video reveals that Claude is trained and served across AWS Trainium, Google TPUs, NVIDIA GPUs and AMD GPUs, and that a system called Mythos ported Claude to AMD in three days, writing almost all the kernels.
Simon Willison released llm-gemini 0.36, adding Google's new Gemini 4 Carbon model to the LLM command-line tool now that it is generally available at $2 per million input and $10 per million output tokens until prices double on December 1st.
Meta AI released Muse Ember, a small, low-cost model aimed at agent workloads such as tool calls, coding subtasks, routing, extraction and long document parsing, priced at $0.10 per million input tokens and $0.50 per million output tokens for shorter prompts.
Anthropic's testing found that telling Claude Fable 5.2 it owns and maintains the codebase raised consolidation of duplicated code from under 10 percent of runs to about 75 percent, versus prompts that implied the code belonged to someone else.
Simon Willison's llm-anthropic 0.31 adds Anthropic's new Fable 5.2 and Mythos 5.2 models; Fable 5.2 costs $8/million input and $40/million output, a 20% discount on Fable 5.1.
Simon Willison used Claude Code to build Beep Locator, a browser-based system that triangulates a mystery beep using multiple devices as microphones and located the culprit to within about 40cm — an old carbon monoxide detector with failing batteries.
Simon Willison highlights Terry Pratchett's idea of "horizontal wealth": spending extra income on the things you already loved rather than on luxury status symbols, using it to justify building a library extension and buying books for undirected "blind research".
Nilay Patel argues AI products aren't good enough to justify their costs, saying the industry must build things that visibly improve people's lives before the backlash eases.
datasette-llm 0.1a10 fixes a bug where the available LLM models were not correctly restricted to those set in configuration, affecting Simon Willison's agent.datasette.io demo site.