Dream-1 launches: 41% of subtasks in 5 minutes, outpacing top LLM agents
gharik · x · 2026-10-20
Dream Labs launched Dream-1, and third-party evaluator XDOF released benchmark results from an early preview on its LLM-focused agentic benchmark: Dream-1 Beta completed a mean 41% of subtasks within its 5-minute time limit, while the best agents — Astra (48% final) and Sol (46% final) — needed 14–15 minutes to reach 41%. Opus finished at 30% and Fable at 26%.
XDOF's verdict: Dream-1 performs well at significantly faster execution speeds. Scores are capped per criterion target, ended runs keep their value, and the export is dated 19 Oct 2026.
More from Models
- Day-one $200 Codex subscriber downgrades for the first time, citing compute shortfall — karmay007 · 2026-10-20
- DeepSeek V4.1 Flash ships a two-stage refusal guardrail, uncensoring effort finds — cephaloform · 2026-10-20
- Meta releases Muse Glimmer, a 30B open-weight agentic model that runs on a single consumer GPU — AIatMeta · 2026-10-20
- Maxwell debuts as 'world's first Code Execution Model', early access open — tri_dao · 2026-10-20
- d1-ultra tops Artificial Analysis Decision Index by a wide margin — JosephJacks_ · 2026-10-20
- Deutsche Bank: Meta's Muse still leads ChatGPT in downloads and DAUs 36 days post-launch — alexandr_wang · 2026-10-20