H2O-Lightning-4B: Apache-2.0 4B Decision Model Tops JevBench Open Leaderboard
pseudotensor1234 · reddit · 2026-10-10
H2O.ai released H2O-Lightning-4B (Apache-2.0), a 4B model for 'decisions API' inference: send a state plus typed questions, get calibrated probabilities from one forward pass with no generated tokens. It scores 72.5 on JevBench, above Jev 1.13's 71.5, topping the open leaderboard. Built on a Qwen3.5-4B fine-tune, it runs on stock vLLM with an open shim (30ms per decision on H100). 12B and 31B versions are coming. Weights and demos (insurance claims triage, web agent, DOOM) are on Hugging Face.
More from Models
- Only 7% of training compute goes to pretraining? repligate probes what the figure means — repligate · 2026-10-10
- New GPT Memory Consolidation Model 'Memory4 Dream' Appears, Appears Built on GPT-6 Luna — lyraxana · 2026-10-10
- Grok Bot now acts as an autonomous X research analyst with daily briefings — FinanceYF5 · 2026-10-10
- Doctors Are Building Board-Style Exams for Medical AI, Starting with Radiology's Last Exam — DrDatta_AIIMS · 2026-10-10
- AI roundup: OpenAI tops 700 math papers, Mistral ships Le Chonk open model — FinanceYF5 · 2026-10-10
- Notes from an NYC AI dinner: agents find most inference perf wins, code review deemed unproductive — paulnovosad · 2026-10-10