← The Wire

Research2026-06-07 · source-backed

Agents' Last Exam: hardest tier passes 2.6% of the time.

Confidence
source-backed
Sources
1
Redaction
passed

Story

ALE is 1,000+ economically meaningful tasks mapped to the US occupational taxonomy across finance, healthcare, legal, engineering, and manufacturing. The hardest tier averages a 2.6% full pass rate. Source: Agents' Last Exam The thesis is explicit: SWE-bench made coding measurable and improvable, ALE wants to do that for the rest of GDP-weighted knowledge work. Pair it with SWE-Marathon and the message is consistent. We measure the easy stuff well and the hard stuff barely at all.

Related stories

Each link below shares sources, entities, or timing with this story.

  1. Shared entities / Shared topic / What happened next / Tension

    Agents' Last Exam: frontier agents pass 2.6% of the hardest economically-valuable tasks.

    Both cover ALE, Last Exam; overlapping topics (agent, exam, tier); picks up the ALE thread on 2026-06-18.

  2. Shared entity: SWE / Shared topic / Earlier coverage / Tension

    SWE-bench Evolving: Opus 4.6 Leads at 79.2%, But Harder Variants Expose Limits

    Both cover SWE; overlapping topics (agent, coding, engineering); earlier SWE coverage from 2026-02-12.

  3. Shared entity: SWE / Shared topic / What happened next / Tension

    Cohere open-sourced a 30B coding model that runs on a single H100. Agentic coding without the frontier bill.

    Both cover SWE; overlapping topics (agent, coding); picks up the SWE thread on 2026-06-10.

  4. Shared entities / Shared topic

    SWE-Marathon: no coding agent clears 19% on ultra-long-horizon tasks.

    Both cover Marathon, SWE; overlapping topics (agent, coding).

  5. Shared entity: SWE / Shared topic / What happened next

    Coding Benchmarks Conflate Model, Retrieval, and Harness

    Both cover SWE; overlapping topics (agent, coding, engineering); picks up the SWE thread on 2026-07-01.

  6. Shared entity: SWE / Shared topic / Earlier coverage / Tension

    Kimi K2.6 Beats GPT-5.4 on SWE-Bench Pro. It's Open-Weight. It Costs 88% Less.

    Both cover SWE; overlapping topics (agent, coding); earlier SWE coverage from 2026-05-05.

  7. Kimi K2.6: A 1-Trillion-Parameter Open-Weights Model Just Beat GPT-5.4 on SWE-Bench Pro

    Both cover SWE; overlapping topics (agent, coding); earlier SWE coverage from 2026-04-21.

  8. An Open-Source Model Just Beat GPT-5.4 and Opus 4.6 on the Hardest Coding Benchmark

    Both cover SWE; overlapping topics (coding, hardest); earlier SWE coverage from 2026-04-20.

Source trail

Entities

Provenance

AI generated
no
Story unit
2026-06-07-agents-last-exam-hardest-tier-passes-2-6-of-the-time
Labels
source-backed, canonical briefing excerpt