Research2026-06-18 · source-backed

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

Issue

2026-06-18

Confidence

source-backed

Sources

Redaction

passed

Story

Built with 250-plus industry experts, ALE evaluates agents on 960 expert-authored, deterministically-scored workflows across 55 industries. Agents average 26% across all tiers but 2.6% on the "last-exam" tier. The kicker: Codex with GPT-5.5 scores 82% on Terminal-Bench and 0% on Last-Exam tasks. The next time someone says agents are "job-ready," this is the number to quote back. (arXiv 2606.05405)

Source trail

arXiv 2606.05405 More from arxiv.org

Entities

ALE Bench Built Codex Exam GPT Last Last Exam Terminal

Provenance

Canonical issue: Ramsay Research Agent — June 18, 2026
AI generated: no
Story unit: 2026-06-18-agents-last-exam-frontier-agents-pass-2-6-of-the-hardest-economically-valuable-tasks
Labels: source-backed, canonical briefing excerpt

Fetching from the wire…

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

Story

Related stories

Codex on GPT-5.5 tops Terminal-Bench 2.1 at 83.4%, but the top Claude config is export-banned.

GPT-5.5 Is Here. It's Fast, It's Expensive, and It Changes How You Prompt.

Wrapping the agent loop in a runtime-guidance DSL lifted skill success 24.1%, tested on Claude Code and Codex directly.

Codex CLI on GPT-5.5 tops Terminal-Bench 2.1 at 83.4%.

GPT-5.5 powers Codex: 82.7% Terminal-Bench 2.0, native Windows, 2x Pro usage through May.

66.5% of Malicious GitHub Issues Get Past Every Guardrail in Cursor, Claude Code, and Codex

Source trail

Entities

Provenance

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

Story

Related stories

Codex on GPT-5.5 tops Terminal-Bench 2.1 at 83.4%, but the top Claude config is export-banned.

GPT-5.5 Is Here. It's Fast, It's Expensive, and It Changes How You Prompt.

Wrapping the agent loop in a runtime-guidance DSL lifted skill success 24.1%, tested on Claude Code and Codex directly.

Codex CLI on GPT-5.5 tops Terminal-Bench 2.1 at 83.4%.

GPT-5.5 powers Codex: 82.7% Terminal-Bench 2.0, native Windows, 2x Pro usage through May.

66.5% of Malicious GitHub Issues Get Past Every Guardrail in Cursor, Claude Code, and Codex

Source trail

Entities

Provenance

Story

Related stories

Codex on GPT-5.5 tops Terminal-Bench 2.1 at 83.4%, but the top Claude config is export-banned.

GPT-5.5 Is Here. It's Fast, It's Expensive, and It Changes How You Prompt.

Wrapping the agent loop in a runtime-guidance DSL lifted skill success 24.1%, tested on Claude Code and Codex directly.

Codex CLI on GPT-5.5 tops Terminal-Bench 2.1 at 83.4%.

GPT-5.5 powers Codex: 82.7% Terminal-Bench 2.0, native Windows, 2x Pro usage through May.

66.5% of Malicious GitHub Issues Get Past Every Guardrail in Cursor, Claude Code, and Codex

Tomorrow's issue, in your inbox.

Source trail

Entities

Provenance

Story

Related stories

Codex on GPT-5.5 tops Terminal-Bench 2.1 at 83.4%, but the top Claude config is export-banned.

GPT-5.5 Is Here. It's Fast, It's Expensive, and It Changes How You Prompt.

Wrapping the agent loop in a runtime-guidance DSL lifted skill success 24.1%, tested on Claude Code and Codex directly.

Codex CLI on GPT-5.5 tops Terminal-Bench 2.1 at 83.4%.

GPT-5.5 powers Codex: 82.7% Terminal-Bench 2.0, native Windows, 2x Pro usage through May.

66.5% of Malicious GitHub Issues Get Past Every Guardrail in Cursor, Claude Code, and Codex

Tomorrow's issue, in your inbox.

Source trail

Entities

Provenance