Google unveils Gemini 4 Argon with 77.9% on DeepSWE, but nobody can use it yet
Ars Technica AI · rss · 2026-10-01
Google announced Gemini 4 Argon, claiming industry-leading performance in coding, knowledge work, and cybersecurity — but the model isn't available to the public yet.
Key facts:
- Already used internally: Argon helped Google save 300 TiB of datacenter memory via fleet-wide telemetry data
- Its agents are migrating Google's C/C++ codebases to Rust, including re2, libgav1, and 800K+ lines of the Fuchsia OS Zircon kernel
- Scores 77.9% on the DeepSWE v1.1 software engineering benchmark, beating GPT-6 Astra, Fable 5.1, and Opus 5.5
- Top score on the economic analysis Vals Index
Ars Technica notes Google had promised Gemini 3.5 Pro for June but spent the summer shipping smaller Flash models before returning to the frontier.
More from Companies & People
- Patrick Collison hosting small AI x biology event with Arc Institute researchers — BrianHie · 2026-10-01
- Google DeepMind Tokyo is hiring for Gemini Live voice tech and multilingual LLM research — AleksandraFaust · 2026-10-01
- Meta claimed Zuck as a "researcher" to snag a $355M tax break on his $4.1B stock compensation — haydenfield · 2026-10-01
- What's the moat for frontier labs like OpenAI and Anthropic anymore? — gizman95 · 2026-10-01
- Paul Graham's 12-year-old essay resurfaces: mean people fail at startups — vaibhavbetter · 2026-10-01
- HF researcher admits a well-maintained monorepo is the superior way to run an AI lab — soldni · 2026-10-01