Sai agent tops OSWorld 2.0 at 79.38%, costs $14.34 per task vs Claude's $24.11
nikola_mr64990 · x · 2026-10-06
- SimularAI's model-agnostic computer-use agent Sai is #1 on OSWorld 2.0 with a 79.38% partial success rate in real-world desktop environments.
- Cost per task: Sai at $14.34 vs Anthropic's Claude Computer Use with Opus 5 at $24.11.
- The team frames it as a new, more reliable and cost-efficient approach to autonomous computer execution.
Related event: SimularAI's Sai Tops OSWorld 2.0 at Lower Cost(2 posts)→
More from coding & agent
- Plainserp: a $0.30/1k-search Google SERP API built for AI agents — Weak_Associate4176 · 2026-10-06
- Local 27B model reads a lease and spots a billing leak — no cloud, no API keys — Acceptable-Object390 · 2026-10-06
- Google's SHIFT Builds Per-Query Multi-Agent Harnesses, Beats 17 Baselines by 7.2 Points — google · 2026-10-06
- Tiny decision model Jev classifies 1,000 papers for $0.08, letting frontier LLMs skip yes/no drudgery — TinfoilTricorn · 2026-10-06
- ChatGPT iOS update quietly adds Codex widgets with lock-screen usage limits — Dimillian · 2026-10-06
- 5 tools for operating AI agents at scale: LangGraph, Port.io, PydanticAI, Langfuse, n8n — Electrical-Stage-346 · 2026-10-06