Gemini 4 Argon writes up to 1M tokens per response, but almost nobody can use it yet
lmoroney · x · 2026-10-01
Google announced Gemini 4 Argon, a frontier model that can output up to 1 million tokens per response (up from 64K), with no size or input-context disclosures. Google claims internal agents use it to migrate C/C++ codebases to Rust — from tens of thousands of lines up to 800K+ lines of the Fuchsia Zircon kernel — and replaced 32K lines of SIMD in libgav1's Rust port, claiming a 2.7x faster memory-safe decoder. Laurence Moroney notes it's rolling out to security teams first, urges caution on vendor benchmarks, and explains how to prepare today for long-running agents.
Related event: Google Launches Frontier Model Gemini 4 Argon(42 posts)→
More from coding & agent
- Cua CEO on why AI agents still screenshot the web — and moving beyond MCP tool definitions — thursdai_pod · 2026-10-01
- OpenAI positions Dots as an orchestrator for engineering across local and cloud environments — Dimillian · 2026-10-01
- Evidentiality framework labels every model claim as given, verified, or generated — early tests look promising — jzesbaugh · 2026-10-01
- A reusable agent prompt for finding Reddit questions your product can actually answer — investigatormaker · 2026-10-01
- Developer review: Muse is good but the agent is slow, dumb, and battery-hungry — ivan_bezdomny · 2026-10-01
- Building a Minimal Coding Agent From Scratch: Architecture Deep-Dive — nik-55 · 2026-10-01