Simular's Agent Sai Tops OSWorld 2.0, Beating GPT-5.6 at Two-Thirds the Cost
xwang_lk · x · 2026-08-28
Simular's computer agent Sai achieved a 73% success rate on OSWorld 2.0, ahead of GPT-5.6 Sol (62.57%, per OpenAI) and Opus 5 (70.57%, per Anthropic), at roughly two-thirds the cost.
OSWorld 2.0, released in June by HKU's XLANG Lab, has 108 long-horizon tasks taking skilled humans over an hour each: cross-source reasoning (receipts scattered across email and expense reports), responding to dynamic changes, following tutorials precisely, and troubleshooting contradictory data.
Sai operates desktop apps and webpages, calls APIs, and writes code like a human — orchestrating a mixture of frontier and specialist models with dedicated perception/action interfaces, running on Simular's remote fleet or personal machines.
Related event: Simular's Sai Agent Tops OSWorld 2.0 at Two-Thirds the Cost(4 posts)→
More from coding & agent
- Animated teardown of XBOW's Chrome full-chain exploit that won Google's first $250K VRP bounty — moyix · 2026-10-15
- pstack's durable core: 10-20 line skills that distill agent best practices into principles — HankYeomans · 2026-10-15
- 'Children of Elon' forum lets AI agents post and teach each other while humans watch — AiJohnAllen · 2026-10-15
- VoidZero open-sources Void: SDK, CLI and platform to deploy Vite apps on Cloudflare — cnakazawa · 2026-10-15
- Running 2 agents on the same data breaks everything — Omnigraph offers a fix — sarahdrinkwater · 2026-10-15
- Evan You presents 'Developer Tooling in the Agentic Era' at ViteConf 2026 — cnakazawa · 2026-10-15