Gemini 3.6 Flash Tested: Higher Pass Rates and Enhanced Program Rebuilds
jyangballin · x · 2026-08-07
In the ProgramBench benchmark, Gemini 3.6 Flash becomes the first Gemini model to fully resolve a task (cmatrix), passing 100% of its behavioral tests. The test requires an agent to rebuild a program from scratch using only the compiled binary and docs, without source code or a decompiler. The average pass rate climbs from 53.6% to 55.7%, and near-perfect rebuilds (>95% of tests passed) increase from 6 to 8 tasks.
Related event: Gemini 3.6 Flash Tops ProgramBench Reverse Engineering Test(3 posts)→
More from Models
- OpenAI stealth-testing mysterious internal model 'Aeon' in Codex desktop app and Dots — cedric_chee · 2026-10-16
- Argon edges out astra as daily driver, but Claude still wins heavy coding work — iruletheworldmo · 2026-10-16
- NVIDIA ships FP4-quantized Qwen3.8-Flash for local use, trending on Hugging Face — nvidia · 2026-10-16
- RoboVista benchmark: Gemini-4 Argon hits 81.9% on robotics understanding — berkeley_ai · 2026-10-16
- Dev reports decision model jev beats gpt-6-luna on quality in real training pipeline — genmon · 2026-10-16
- Google won't publish Argon 4 limits in AI Pro—users compare experiences — Consistent-Text-2662 · 2026-10-16