DATUM

llm-gemini 0.34 adds Gemini 3.8 Flash with three thinking levels

single source · 1 articles · release · confidence: high · first seen 2026-09-02 16:39 UTC

Simon Willison released llm-gemini 0.34, adding support for Google's new Gemini 3.8 Flash model with three 'thinking levels' (low, medium, high) that control how much computation the model uses before answering. The update also fixes async responses failing to record the resolved model version. Google released Gemini 3.8 Flash today; a 'Cyber' variant is available only to 'trusted defenders'. Willison reports generating a working HTML page in 13 seconds for 1.8 cents and using the model to add sandboxed HTML rendering to his markdown-svg-renderer tool.

What this means for you

If you use Simon Willison's LLM command-line tool, upgrade to llm-gemini 0.34 to get Gemini 3.8 Flash and the async-response fix. The 1.8-cent, 13-second HTML build is one developer's test, not a benchmark.

Key facts

  • ·llm-gemini 0.34 released by Simon Willison. source
  • ·Adds support for Google's Gemini 3.8 Flash with low, medium and high thinking levels. source
  • ·Fixes async responses failing to record resolved model version (issue #146). source
  • ·Google released Gemini 3.8 Flash today; a Cyber variant is limited to trusted defenders. source
  • ·Willison generated an HTML page in 13 seconds at a cost of 1.8 cents. source

What the sources say

  • Simon Willison (press) — Release note for llm-gemini 0.34 with model support details and hands-on cost and speed test.

Sources

The original reporting. Follow these — they did the work.

← the wire