Acellera tests 7 LLM+harness combos on drug discovery: one RTX 5090 holds up
gdefabritiis · x · 2026-09-10
Acellera founder Gianni De Fabritiis reports the company burns 1B tokens/day via agentic workflows, expected to grow 10x — making frontier-only usage unaffordable. He tested how far self-hosted models, including single-GPU ones, can go.
The experiment: 7 LLM + harness combinations (Claude, ChatGPT, PlayMolecule® AI) on a demanding IL-23 (IL23R-p19, psoriasis) drug discovery task — retrieving literature and structural data to produce a measurable target product profile and small-molecule strategy. Several surprises, including what a single RTX 5090 could produce. Blog and comparative reports are public.
Related event: Acellera Tests Local LLM Drug Discovery on a Single RTX 5090(2 posts)→
More from Infra
- Matt Barrie burned 4B tokens in a day, cut his bill 500-fold, and now worries about $5T in debt — gaganghotra_ · 2026-09-10
- Analyst: DeepSeek's latest change is a big win for token efficiency, moving toward OpenAI's regime — teortaxesTex · 2026-09-10
- Mac mini tested: local 35B runtime hits Haiku-level scores but falls short for agents — PawelHuryn · 2026-09-10
- Miles ships Day-0 RL support for DeepSeek-V4.1-Flash with KL held at 0.0012–0.0017 — ying11231 · 2026-09-10
- The data center is a symbol: why debunked claims about AI infrastructure still spread — ShakeelHashim · 2026-09-10
- Kimi K3 lands on RunPod: 2.8T params, 1M context, $3/$15 per 1M tokens — Kimi_Moonshot · 2026-09-10