Gemini 4 Hydrogen scores 13.5% on SWE-sweep, 2.3x GPT-6.1 Terra, but leads unevenly
ChrisGPT · x · 2026-11-13
ChrisGPT shares Gemini 4 Hydrogen benchmark results (Nov 2026): 13.5% on SWE-sweep vs 5.8% for GPT-6.1 Terra and 3.1% for Claude Sonnet 5.5 — roughly 2.3x Terra. Hydrogen leads 7 of 10 rows, including Agents' Last Exam v1.1 (71.6%), AutomationBench-AA (76.1%), Vending-Bench 2 ($14,549 simulated balance) and HLE-Diamond (58.8%). But Sonnet 5.5 wins Terminal-Bench 4.1 (63.9% vs 57.0%), the OSWorld-2.1 offline subset (83.9%) and Terminal-Bench-Science (59.9%), so gains don't transfer uniformly across terminal, computer-use and science tasks. The author also flags grading weaknesses: Epoch found serious issues in Agents' Last Exam v1.1 public tasks.
More from Models
- Gemini 4 Hydrogen Rises to #4 in Knowledge Work on Agent Arena, Beating Three Higher-Ranked Rivals — arena · 2026-11-13
- Livestream Leak: Upcoming Cyber Model Hits 77% on ExploitGym vs Rivals' 30-50% — Kyrannio · 2026-11-13
- Leak: Gemini 4 Hydrogen surfaces — 2.5x faster than 3.8 Flash, beats Argon, cheaper — Electronic-Weekend66 · 2026-11-13
- Gemini UI screenshot shows 4 Hydrogen / Argon / Carbon model tiers in picker — Then_Flounder6906 · 2026-11-13
- Ditto rolls out uncensored inference to all users — bittingthembits · 2026-11-13
- Google says Gemini 4 Hydrogen expands query fan-out and the range of sites Search links to — OnlineInference · 2026-11-13