COMPANY

SemiAnalysis

companytopic-noteresearch

Overview

SemiAnalysis is a semiconductor and AI-compute research firm whose coverage of frontier-lab economics, data-center buildouts, and inference stacks has made it a durable primary source in the AI Digest corpus. Its first dedicated appearance as a topic note is for the 2026-11-06 publication of GDPx — a 206-task agent benchmark across 11 domains chosen to be representative of knowledge-work economy value, with no agent scoring above 12.0% across the full suite.

Timeline

  • 2026-11-06-AI-Digest — SemiAnalysis published GDPx, a new agent benchmark covering 206 expert-authored tasks across 11 domains chosen to be representative of knowledge-work economy value; no agent evaluated scored above 12.0% across the full suite, with Opus 5.5 the reported leader at the ceiling. The release is explicitly framed as “evaluate agents on real work, not coding leaderboards”. Scope clarification worth carrying: GDPx is not a “frontier agents fail” result — it’s “frontier agents fail on tasks SemiAnalysis specifically chose to be hard”. Pair against the Aider polyglot top-5 (gpt-5 high at 88.0% cross-language coding) — the two benchmarks measure different things, not conflicting pictures of the same agent. Read with today’s Harness-S paper: if base-harness bottlenecks are a first-order driver of real-work ceilings, then a 12% GDPx ceiling and an 88% polyglot ceiling can both be stable pictures of the same model behaviour. Log against MOC - Agentic Coding.

Key Developments

  1. GDPx Benchmark Posts a 12% Agent Ceiling Across 206 Tasks and 11 Domains (November 6, 2026): SemiAnalysis published GDPx as a knowledge-work-weighted benchmark with no agent above 12.0% across the full suite; Opus 5.5 leads at the ceiling. Explicit framing: “evaluate agents on real work, not coding leaderboards”. Load-bearing scope clarification: this is 12% on tasks SemiAnalysis chose to be hard, not a frontier-agents-fail-everywhere result — paired against gpt-5 (high) at 88.0% on Aider polyglot, the two benchmarks measure different things. Reading with the same-cycle Harness-S paper (which recovers successful trajectories for agent RL by routing through alternative execution harnesses when base-harness rollouts fail) suggests a 12% GDPx ceiling and an 88% polyglot ceiling can both be stable pictures of the same model behaviour where the harness is a load-bearing variable. Covered in 2026-11-06-AI-Digest.

See also: MOC - Agentic Coding, MOC - AI Infrastructure.