Parallel Launches Agentic Search Leaderboard: GPT-6 Tops at 70.3, DeepSeek Gains Most
rickasaurus · x · 2026-09-19
Parallel (parallel.ai) introduced the Search Capability Leaderboard, benchmarking how well top models orchestrate web searches and synthesize correct answers, covering answer quality, cost, and the lift from search access.
Key takeaways:
- GPT-6 (Astra): highest Search Intelligence score (70.3), best for complex reasoning and maximum accuracy.
- GPT-5.6 Luna: most cost-efficient ($23.0 per 1K tasks), 13× cheaper than the top model.
- DeepSeek V4.1 Flash: largest lift from search (+45.0), plus Silver in efficiency and Bronze in intelligence.
Alongside the leaderboard, Parallel also shipped Parallel Search Fast, a fast, cheap web search API designed for search-heavy agent workflows.
More from coding & agent
- OpenClaw launches Multiplayer Mode to kill the 'meat proxy' in agent teamwork — heyneighbor · 2026-09-19
- Computer-use agents aren't GUI agents: best agents will code, call tools and use the CLI — DhruvBatra_ · 2026-09-19
- Muse agent books real flights end-to-end with zero errors, user reports — alexandr_wang · 2026-09-19
- Browser use is a commodity with no moat, argues dev as top sites may build native agent tools — tedddyoweh · 2026-09-19
- Open-source demo: TypeSafe Jev as a typed router inside a Flue agent on Cloudflare — irvinebroque · 2026-09-19
- Flue 2.1 ships per-tool timeouts, MCP tool annotations and trace budgets — irvinebroque · 2026-09-19