Perplexity releases local-first Portable Computer, 27B model beats open-source harnesses
perplexity_ai · x · 2026-08-26
Perplexity introduced Portable Computer, a local-first agent framework designed for private and cost-effective work.
Performance:
- On web research tasks (1,266 BrowseComp tasks), Computer achieved 66.7% accuracy, outperforming Pi (50.2%) and Hermes (43.9%).
- On a real-world knowledge work benchmark, their 27B model scored 82.6%, while the post-trained PPLX 27B reached 85.4%, both beating existing open-source harnesses.
Architecture:
- Model and harness are co-designed (small models fail in harnesses built for frontier models).
- Features a minimal system prompt, on-demand skills, and compact CLI tools as connectors instead of MCP servers.
- Includes self-verification and an always-on sandbox.
Related event: Perplexity Launches Fully Local Agent Stack Portable Computer on DGX Spark(17 posts)→
More from coding & agent
- Claude/Codex built my Unity game, but visual testing is hard — PerspectivePersonal · 2026-08-26
- Agent worked in demo, broke in prod. How to build trust? — Prod_whiz · 2026-08-26
- Train a private SLM on iMessages locally with MLX — romainhuet · 2026-08-26
- Dev builds a playable Game Boy Advance shooter 'Sunshard' with Codex — romainhuet · 2026-08-26
- Claude Reflect System: Enables permanent learning from corrections — tom_doerr · 2026-08-26
- Bug Hunt Bench: Benchmarking frontier coding models on real planted bugs — PawelHuryn · 2026-08-26