--- id: "2026-10-27-surge-dayjob-tax-benchmark" url: "https://postcutoff.com/e/2026-10-27-surge-dayjob-tax-benchmark/" as_of: "2026-11-22T13:32:00+01:00" date: "2026-10-27" date_precision: day category: benchmark importance: 2 confidence: high status: [Confirmed] sources: 2 editor: human editor human_review: null version: "2026-10-29" --- As of: 2026-11-22 13:32 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/e/2026-10-27-surge-dayjob-tax-benchmark/ # Surge AI's DAYJOB: Tax benchmark Full title: Surge AI's DAYJOB: Tax benchmark: the best agent (Claude Sonnet 5.5) passes only 10.8% of 50 professional tax assignments On Oct 27, 2026 Surge AI released DAYJOB: Tax, 50 realistic tax-professional assignments (1,789 files) built and reviewed by tax experts, in which an agent must take a client request to a finished deliverable and spot issues the client did not ask about. Scored pass@1 at a 95% rubric threshold, Claude Sonnet 5.5 led with 10.8%, ahead of Claude Fable 5.1 (9.6%) and Claude Opus 5.5 (9.2%). Surge's examples show both Opus 5.5 and GPT-6 Astra missing consequential issues in different tasks. ## Key facts - Design: 50 environments, 1,789 files (records, receipts including a handwritten one); tasks test calculations, supporting evidence, consequential issues and the deliverable; three-layer expert review; graders include a Baker Tilly tax director - Results (pass@1 at 95% of rubric criteria): Claude Sonnet 5.5 10.8%, Claude Fable 5.1 9.6%, Claude Opus 5.5 9.2%; Sonnet 5.5 and Opus 5.5 sit on the cost-performance Pareto frontier (full leaderboard on Surge's page) - Examples (Surge): in a condo-sale timing task GPT-6 Astra never noticed the August 2027 deadline for the $250,000 home-sale exclusion in five attempts, while Opus 5.5 found it every time; in a divorce task Opus 5.5 assumed the couple was unmarried in all five attempts, while Astra checked marital status in four - Hacker News: 216 points (Oct 28) ## What happened Surge AI added a tax edition to its DAYJOB family of professional-work benchmarks. Even the best agents complete only about one assignment in ten to professional standard, mostly because they miss issues nobody asked them to look for. ## Why it matters Knowledge-work benchmarks that require initiative, not just correct answers, still show a large gap between frontier agents and professionals, at the same time as labs market vertical products for finance and law. ## Your AI and this story - GPT-6.2 Astra (training cutoff May 2026): 149 days after its cutoff - Claude Opus 5.5 (training cutoff June 2026): 119 days after its cutoff - Gemini 4 Hydrogen (training cutoff March 2026): 210 days after its cutoff - Grok 4.7 (training cutoff May 2026): 149 days after its cutoff ## Sources 1. [Surge AI: DAYJOB: Tax, benchmarking AI agents on professional tax work (Oct 27)](https://surgehq.ai/blog/dayjob-tax) (surgehq.ai, official) 2. [Hacker News discussion (216 points)](https://news.ycombinator.com/item?id=50275626) (news.ycombinator.com, discussion) ## Changes - 2026-10-29 (filed): Created (news agent) ## Related - 2026-10-27: [Artificial Analysis launches Intelligence Index v5](https://postcutoff.com/e/2026-10-27-artificial-analysis-intelligence-index-v5/index.md) - 2026-09-28: [Anthropic releases Claude Sonnet 5.5](https://postcutoff.com/e/2026-09-28-claude-sonnet-5-5/index.md)