Argon solves ~half of 74 real ML engineering tasks on Google RE-Bench, vs 27% for 3.7 Flash
HaydnBelfield · x · 2026-10-14
Part 7/9 of DeepMind's report thread gives benchmark results: on Google RE-Bench's 74 realistic ML engineering tasks, Gemini 4 Argon solves roughly half of completed runs — a big step up from Gemini 3.7 Flash (27%), but well below the near-perfect score needed for full automation of representative day-to-day ML R&D engineering.
Related event: Google launches Gemini 4 Argon with 217-page safety report(22 posts)→
More from Models
- GPT-6 Astra Reads Only 46% of VINs Correctly vs Sonnet 5's 86%: Routing Pays Off — arnabdotorg · 2026-10-15
- VLMs May Turn Clearly Superhuman in Vision Within a Year or Two, Researcher Predicts — arnabdotorg · 2026-10-15
- Gemini 4 Argon Trails GPT 6 Astra by Just 25 Points on Physical Agents Index — qinzytech · 2026-10-15
- GPT-6-Astra scores 98% on unpublished theoretical physics problems, catching expert errors — sandersted · 2026-10-15
- Devs report quality degradation when Kimi K3 is served via third-party providers — casper_hansen_ · 2026-10-15
- Gartner's 2026 GenAI Hype Cycle: Agentic AI still climbing, AGI 10+ years out — luisdans · 2026-10-15