Skip to main content

Agents Last Exam (ALE-Bench)

A benchmark for agentic professional workflows with verifiable success criteria, reporting pass rates and partial scores for model plus agent-harness rows.

How BenchLM shows ALE-Bench

BenchLM mirrors the Agents Last Exam full/overall split from the public leaderboard API. The snapshot reports pass rate and partial average score across roughly 149 professional workflow tasks for model plus agent-harness rows.

ALE-Bench is display only on BenchLM. Its rows combine a base model with an agent harness such as Codex, OpenClaw, Claude Code, Droid, Cursor CLI, or Gemini CLI, so BenchLM keeps the table separate from model-only rankings.

27 harness rows149 ALE-V1 tasksFull/overall splitOfficial API snapshotDisplay only

Pass rate on ALE-Bench — June 2026 API snapshot

BenchLM mirrors the published pass rate view for ALE-Bench. codex / gpt-5-5 leads the public snapshot at 26.2% , followed by ale_claw / gpt-5-5 (24.2%) and openclaw / gpt-5-5 (22.8%). BenchLM does not use these results to rank models overall.

27 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated June 2026 API snapshot

The published ALE-Bench snapshot is tightly clustered at the top: codex / gpt-5-5 sits at 26.2%, while the third row is only 3.4 points behind. The broader top-10 spread is 10.1 points, so the benchmark still separates strong models even when the leaders cluster.

27 models have been evaluated on ALE-Bench. The benchmark falls in the External benchmark mirrors category. BenchLM tracks this category separately from its weighted global scoring system, so these results are best compared on the dedicated Korean benchmark views. ALE-Bench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About ALE-Bench

Year

2026

Tasks

Approximately 149 ALE-V1 professional workflow tasks

Format

Pass rate and partial-credit score

Difficulty

Real-world agentic workflows

BenchLM mirrors the public Agents Last Exam full/overall leaderboard API as ALE-Bench. Rows combine base models with agent harnesses such as Codex, OpenClaw, Claude Code, Droid, Cursor CLI, and Gemini CLI, so the table remains display-only.

BenchLM freshness & provenance

Version

ALE-Bench 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Pass rate table (27 models)

1
codex / gpt-5-5codex/gpt-5-5
26.2%
2
ale_claw / gpt-5-5ale_claw/gpt-5-5
24.2%
3
openclaw / gpt-5-5openclaw/gpt-5-5
22.8%
4
cursor_cli / gpt-5-5cursor_cli/gpt-5-5
22.5%
5
openclaw / gpt-5-4openclaw/gpt-5-4
22.3%
6
cursor_cli / composer-2-5cursor_cli/composer-2-5
22.1%
7
cursor_cli (thinking-high) / claude-opus-4-7cursor_cli:thinking-high/claude-opus-4-7
21.5%
8
droid / gpt-5-5droid/gpt-5-5
20.1%
9
ale_claw / claude-opus-4-7ale_claw/claude-opus-4-7
19.5%
10
claude_code / claude-opus-4-8claude_code/claude-opus-4-8
16.1%
11
openclaw / claude-opus-4-7openclaw/claude-opus-4-7
16.1%
12
openclaw / gemini-3-1-pro-previewopenclaw/gemini-3-1-pro-preview
15.8%
13
openclaw / claude-opus-4-6openclaw/claude-opus-4-6
15.1%
14
claude_code / claude-opus-4-7claude_code/claude-opus-4-7
14.1%
15
droid / claude-opus-4-7droid/claude-opus-4-7
13.8%
16
openclaw / deepseek-v4-proopenclaw/deepseek-v4-pro
13.0%
17
ale_claw / gpt-5-4ale_claw/gpt-5-4
12.8%
18
openclaw / glm-5-1openclaw/glm-5-1
12.4%
19
openclaw / kimi-k2-6openclaw/kimi-k2-6
9.4%
20
openclaw / qwen3-6-plusopenclaw/qwen3-6-plus
8.7%
21
openclaw / mimo-v2-5openclaw/mimo-v2-5
8.7%
22
codex / gpt-5-4codex/gpt-5-4
7.4%
23
grok_cli / grok-4-3grok_cli/grok-4-3
6.7%
24
openclaw / minimax-m2-7openclaw/minimax-m2-7
6.0%
25
grok_cli / grok-3grok_cli/grok-3
4.7%
26
openclaw / grok-4-3openclaw/grok-4-3
4.4%
27
gemini_cli / gemini-3-1-pro-previewgemini_cli/gemini-3-1-pro-preview
4.0%

FAQ

What does ALE-Bench measure?

A benchmark for agentic professional workflows with verifiable success criteria, reporting pass rates and partial scores for model plus agent-harness rows.

Which model leads the published ALE-Bench snapshot?

codex / gpt-5-5 currently leads the published ALE-Bench snapshot with 26.2% pass rate. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on ALE-Bench?

27 AI models are included in BenchLM's mirrored ALE-Bench snapshot, based on the public leaderboard captured on June 2026 API snapshot.

Last updated: June 2026 API snapshot · mirrored from the public benchmark leaderboard

The AI models change fast. We track them for you.

For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.