Gemini 4 Hydrogen posts 13.5% on SWE-sweep, an standout result for a small-class model
marktenenholtz · x · 2026-11-13
marktenenholtz highlights Google DeepMind's Gemini 4 Hydrogen scoring 13.5% on SWE-sweep — remarkable for a small-class model (vs Claude Sonnet 5.5 at 3.1% and GPT-6.1 Terra at 5.8%). The benchmark table shows it leading on Agents' Last Exam v1.1 (71.6%), AutomationBench-AA (76.1%), Vending-Bench 2 ($14,549), MKA (53.4%), Chartography (66.6%) and HLE-Diamond (58.8%), though it trails Sonnet 5.5 on Terminal-Bench 4.1 (57.0% vs 63.9%) and OSWorld-2.1 (77.6% vs 83.9%).
Related event: Gemini 4 Hydrogen Benchmarks Leak with 13.5% on SWE-sweep(3 posts)→
More from Models
- Unverified claim: hidden GPU fingerprints show top Chinese labs trained on banned Blackwell chips — cis_female · 2026-11-13
- Chinese major models reportedly trained on Blackwell-class chips, though scale disappoints — teortaxesTex · 2026-11-13
- One user burned $4,988 of API-priced tokens on a $200 Claude subscription last month — jonathan_wilke · 2026-11-13
- Gemini 4 Hydrogen tops Google models on AA-Robotics v0.1, ranking fourth overall — ArtificialAnlys · 2026-11-13
- 4,689-trial agent test: DeepSeek V4 argues its way into failures, Gemini 3.8 Flash fails silently — Kindly-Bag763 · 2026-11-13
- Analyst: DeepSeek V4's Flash variant is winning, and a hyper-efficient 2B V5 could be the real play — teortaxesTex · 2026-11-13