Agents Last Exam (ALE-Bench)

We show this table for reference; we do not rank on it.

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

Pass rate on ALE-Bench — June 2026 API snapshot

We mirror the published pass rate view for ALE-Bench. claude_code (thinking-max) / claude-opus-5-5 leads the public snapshot at 38.2%, followed by codex (reasoning-max) / gpt-6-astra (34.2%) and codex (reasoning-xhigh) / gpt-6-astra (32.2%). We do not use these results to rank models overall.

claude_code (thinking-max) / claude-opus-5-5

claude_code

38.2%

codex (reasoning-max) / gpt-6-astra

codex

34.2%

codex (reasoning-xhigh) / gpt-6-astra

codex

32.2%

87 modelsAgenticCurrentDisplay onlyUpdated June 2026 API snapshot

Pass rate table (87 models)

Pass rate results for ALE-Bench
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	claude_code (thinking-max) / claude-opus-5-5claude_codeclaude_code	38.2%	Not reported	Unknown
2	codex (reasoning-max) / gpt-6-astracodexcodex	34.2%	Not reported	Unknown
3	codex (reasoning-xhigh) / gpt-6-astracodexcodex	32.2%	Not reported	Unknown
4	codex (reasoning-medium) / gpt-6-astracodexcodex	32.2%	Not reported	Unknown
5	codex (reasoning-xhigh) / gpt-6-solcodexcodex	32.2%	Not reported	Unknown
6	claude_code (thinking-high) / claude-opus-5claude_codeclaude_code	32.2%	Not reported	Unknown
7	codex (reasoning-xhigh) / muse-spark-1-3codexcodex	32.2%	Not reported	Unknown
8	codex (reasoning-max) / gpt-6-solcodexcodex	31.6%	Not reported	Unknown
9	codex (reasoning-high) / gpt-6-astracodexcodex	31.6%	Not reported	Unknown
10	codex (reasoning-high) / gpt-6-solcodexcodex	31.6%	Not reported	Unknown
11	claude_code (thinking-max) / claude-opus-5claude_codeclaude_code	30.9%	Not reported	Unknown
12	codex (reasoning-xhigh) / GPT-5.6-Solcodexcodex	30.6%	Not reported	Unknown
13	codex (reasoning-high) / GPT-5.6-Solcodexcodex	30.6%	Not reported	Unknown
14	claude_code (thinking-xhigh) / claude-opus-5claude_codeclaude_code	30.3%	Not reported	Unknown
15	codex (reasoning-xhigh) / GPT-5.6-Lunacodexcodex	30.3%	Not reported	Unknown
16	codex (reasoning-max) / muse-spark-1-3codexcodex	29.6%	Not reported	Unknown
17	claude_code (reasoning-high) / minimax-m3-1-flashclaude_codeclaude_code	29.6%	Not reported	Unknown
18	codex (reasoning-low) / gpt-6-astracodexcodex	29.6%	Not reported	Unknown
19	codex (reasoning-max) / GPT-5.6-Solcodexcodex	29.6%	Not reported	Unknown
20	codex (reasoning-medium) / GPT-5.6-Solcodexcodex	29.5%	Not reported	Unknown
21	codex (reasoning-medium) / gpt-6-solcodexcodex	28.9%	Not reported	Unknown
22	claude_code (thinking-medium) / claude-opus-5claude_codeclaude_code	28.9%	Not reported	Unknown
23	codex (reasoning-max) / GPT-5.6-Lunacodexcodex	28.9%	Not reported	Unknown
24	codex (reasoning-none) / gpt-6-astracodexcodex	28.3%	Not reported	Unknown
25	kimi_code (thinking-max) / kimi-k3kimi_codekimi_code	28.3%	Not reported	Unknown
26	codex (reasoning-max) / GPT-5.6-Terracodexcodex	28.0%	Not reported	Unknown
27	codex (reasoning-low) / gpt-6-solcodexcodex	27.6%	Not reported	Unknown
28	claude_code (thinking-low) / claude-opus-5claude_codeclaude_code	27.6%	Not reported	Unknown
29	codex (reasoning-xhigh) / GPT-5.6-Terracodexcodex	27.6%	Not reported	Unknown
30	codex (reasoning-high) / minimax-m3-1-flashcodexcodex	27.0%	Not reported	Unknown
31	claude_code (thinking-xhigh) / qwen-3-8-maxclaude_codeclaude_code	27.0%	Not reported	Unknown
32	grok_build (reasoning-high) / grok-4-5grok_buildgrok_build	27.0%	Not reported	Unknown
33	claude_code (thinking-max) / kimi-k3claude_codeclaude_code	27.0%	Not reported	Unknown
34	claude_code (thinking-max) / claude-opus-4-8claude_codeclaude_code	27.0%	Not reported	Unknown
35	codex (reasoning-xhigh) / gpt-5-5codexcodex	26.6%	Not reported	Unknown
36	codex (reasoning-high) / GPT-5.6-Terracodexcodex	26.0%	Not reported	Unknown
37	claude_code (reasoning-xhigh) / anthropic-claude-fable-5claude_codeclaude_code	25.7%	Not reported	Unknown
38	codex (reasoning-max) / gpt-6-lunacodexcodex	25.0%	Not reported	Unknown
39	codex / gpt-5-5codexcodex	24.5%	Not reported	Unknown
40	codex (reasoning-xhigh) / gpt-6-lunacodexcodex	24.3%	Not reported	Unknown
41	codex (reasoning-high) / GPT-5.6-Lunacodexcodex	24.3%	Not reported	Unknown
42	codex (reasoning-low) / GPT-5.6-Solcodexcodex	23.9%	Not reported	Unknown
43	codex (reasoning-none) / gpt-6-solcodexcodex	23.7%	Not reported	Unknown
44	ale_claw / gpt-5-5ale_clawale_claw	23.0%	Not reported	Unknown
45	codex (reasoning-medium) / GPT-5.6-Terracodexcodex	22.7%	Not reported	Unknown
46	claude_code (thinking-xhigh) / claude-opus-4-8claude_codeclaude_code	22.4%	Not reported	Unknown
47	claude_code / anthropic-claude-fable-5claude_codeclaude_code	22.4%	Not reported	Unknown
48	claude_code (thinking-high) / claude-opus-4-8claude_codeclaude_code	22.4%	Not reported	Unknown
49	agnes_harness / Agnes-2.5-Pro-Betaagnes_harnessagnes_harness	21.7%	Not reported	Unknown
50	openclaw / gpt-5-5openclawopenclaw	21.7%	Not reported	Unknown
51	cursor_cli / gpt-5-5cursor_clicursor_cli	21.4%	Not reported	Unknown
52	codex (reasoning-high) / gpt-6-lunacodexcodex	21.1%	Not reported	Unknown
53	cursor_cli (thinking-high) / claude-opus-4-7cursor_clicursor_cli	21.1%	Not reported	Unknown
54	openclaw / gpt-5-4openclawopenclaw	20.5%	Not reported	Unknown
55	codex (reasoning-medium) / gpt-6-lunacodexcodex	20.4%	Not reported	Unknown
56	claude_code (reasoning-max) / qwen-3-8-27bclaude_codeclaude_code	20.4%	Not reported	Unknown
57	claude_code (max) / glm-5-2claude_codeclaude_code	20.4%	Not reported	Unknown
58	codex (reasoning-low) / GPT-5.6-Terracodexcodex	20.4%	Not reported	Unknown
59	cursor_cli / composer-2-5cursor_clicursor_cli	20.4%	Not reported	Unknown
60	droid / gpt-5-5droiddroid	19.7%	Not reported	Unknown
61	claude_code / ark-0614cclaude_codeclaude_code	19.1%	Not reported	Unknown
62	ale_claw / claude-opus-4-7ale_clawale_claw	18.4%	Not reported	Unknown
63	codex (reasoning-medium) / gpt-5-5codexcodex	18.4%	Not reported	Unknown
64	codex (reasoning-medium) / GPT-5.6-Lunacodexcodex	17.1%	Not reported	Unknown
65	codex (reasoning-low) / gpt-5-5codexcodex	17.1%	Not reported	Unknown
66	claude_code / claude-opus-4-8claude_codeclaude_code	16.4%	Not reported	Unknown
67	codex (reasoning-low) / gpt-6-lunacodexcodex	16.4%	Not reported	Unknown
68	gemini_cli / gemini-3-1-pro-previewgemini_cligemini_cli	16.4%	Not reported	Unknown
69	openclaw / claude-opus-4-7openclawopenclaw	15.8%	Not reported	Unknown
70	openclaw / gemini-3-1-pro-previewopenclawopenclaw	14.1%	Not reported	Unknown
71	codex (reasoning-none) / gpt-6-lunacodexcodex	13.8%	Not reported	Unknown
72	claude_code / claude-opus-4-7claude_codeclaude_code	13.8%	Not reported	Unknown
73	droid / claude-opus-4-7droiddroid	13.5%	Not reported	Unknown
74	openclaw_cli / ark-0614copenclaw_cliopenclaw_cli	12.5%	Not reported	Unknown
75	openclaw / deepseek-v4-proopenclawopenclaw	12.4%	Not reported	Unknown
76	openclaw / qwen-qwen3-7-maxopenclawopenclaw	11.8%	Not reported	Unknown
77	codex (reasoning-low) / GPT-5.6-Lunacodexcodex	11.8%	Not reported	Unknown
78	ale_claw / gpt-5-4ale_clawale_claw	11.8%	Not reported	Unknown
79	openclaw / glm-5-1openclawopenclaw	11.5%	Not reported	Unknown
80	openclaw / kimi-k2-6openclawopenclaw	9.2%	Not reported	Unknown
81	openclaw / qwen3-6-plusopenclawopenclaw	8.6%	Not reported	Unknown
82	openclaw / mimo-v2-5openclawopenclaw	8.6%	Not reported	Unknown
83	grok_cli / grok-4-3grok_cligrok_cli	7.2%	Not reported	Unknown
84	codex / gpt-5-4codexcodex	7.2%	Not reported	Unknown
85	openclaw / minimax-m2-7openclawopenclaw	5.9%	Not reported	Unknown
86	grok_cli / grok-3grok_cligrok_cli	4.6%	Not reported	Unknown
87	openclaw / grok-4-3openclawopenclaw	4.3%	Not reported	Unknown

How ALE-Bench is shown here

BenchLM mirrors the Agents Last Exam full split from the public leaderboard API. The snapshot reports pass rate, partial average score, cost, token, and duration metadata across 152 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.

The Agent Showdown analysis adds domain and failure-mode context across 13 top-level domains. It also notes that Claude Code plus Fable 5 may include fallback to Opus 4.8 on refused tasks, so BenchLM preserves the official mixed-system row label instead of treating it as a pure base-model score.

Snapshot

87 harness rows152 ALE-V1 tasks13 domainsFull splitOfficial API snapshotDisplay only

Agents Last Exam leaderboard Leaderboard API Agent Showdown analysis GitHub repository

The published ALE-Bench snapshot places claude_code (thinking-max) / claude-opus-5-5 first at 38.2%. The third row is 6.0 points behind. The broader top-10 range is 6.6 points, so many of the published results sit in a relatively narrow band.

87 models have been evaluated on ALE-Bench. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. 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

152 ALE-V1 professional workflow tasks across 13 top-level domains

Format

Pass rate, partial-credit score, cost, token, and duration metadata

Difficulty

Real-world agentic workflows

BenchLM mirrors the public Agents Last Exam full leaderboard API as ALE-Bench and links the June 2026 Agent Showdown analysis for domain, cost, speed, and failure-mode context. 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. The source notes that Claude Code plus Fable 5 may include upstream fallback to Opus 4.8 on refused tasks.

Agents Last Exam Public benchmark source

Freshness and 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.

Questions

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?

claude_code (thinking-max) / claude-opus-5-5 currently leads the published ALE-Bench snapshot with 38.2% 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?

The June 2026 API snapshot snapshot contains 87 AI models.

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

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.