Benchmark Almanac

Agents' Last Exam v1.1

GPT-6.2 Astra 47.5% leads since November 2026
GPT-6.2 Astra 47.5% leads since November 2026
In use
Version 1.1

Agents complete 122 public long-horizon professional workflow tasks across 55 sub-industries. Deterministic graders check outcomes against hidden references.

Released October 26th, 2026 as a new public split. ALE rotates fresh tasks into the public pool about every six months, and v1.0 now sits on a legacy board.

Released
October 2026
Kind
Tasks from real jobs
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text, Documents → Actions, Documents, Text
Interaction
Agentic
Environment
Terminal, APIs and tools
Run by
Anyone
In views
Agents, Kept fresh
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last in a launch post
November 2026
Last model added
November 2026
Scores refreshed
November 18th, 2026

In the authors’ words

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long horizon, economically valuable, real world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 sub fields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is below 1%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP relevant impact.

Abstract of Agents' Last Exam (arXiv:2606.05405, CC0)
19
Entries on its own leaderboard, 18 models we track
8
Launch posts list it, from 5 labs
2
System cards report it, from 1 lab

GPT-6.2 Astra leads with 47.5% GPT-6.2 Astra leads with 47.5%

Best score per entry from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

GPT-6.2 Astra leads with 47.5%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model. Striped bars are claims, and dotted bars are reruns by Epoch AI or another board.

  1. 1 GPT-6.2 Astra xhigh 47.5%
  2. 2 Claude Fable 5.2 max 46.9%
  3. 3 qwen-4 xhigh 46.7%
  4. 4 Gemini 4 Carbon high 45.9%
  5. 5 Claude Opus 5.5 max 43.4%
  6. 6 Kimi K3.1 Pro max 43.4%
  7. 7 Gemini 4 Hydrogen xhigh 43.4%
  8. 8 GPT-6 Astra max 42.6%
  9. 9 GPT-6.1 Terra max 41.8%
  10. 10 GPT-6.2 Sol xhigh 41.8%
  11. 11 Tencent Hy4 high 41%
  12. 12 Gemini 4 Argon high 40.2%
  13. 13 Muse Wonder high 38.5%
  14. 14 DeepSeek V4.1 Flash 1017 high 36.1%
  15. 15 Kimi K3 max 35.3%
  16. 16 GPT-5.6 Sol max 30.3%
  17. 17 MiniMax-M3.1-Flash max 29.5%
  18. 18 GLM-5.3 max 26.2%
  19. 19 GPT-5.6 Luna none 5.74%

Source: Official Agents' Last Exam v1.1 leaderboard

#EntryScoreSettingDateSource
1 GPT-6.2 Astra 47.5% xhigh, Codex Nov 2026 Official leaderboard
2 Claude Fable 5.2 46.9% max, Claude Code Nov 2026 Official leaderboard
3 qwen-4 46.7% xhigh, Codex Official leaderboard
4 Gemini 4 Carbon 45.9% high, ALE-Claw Oct 2026 Official leaderboard
5 Claude Opus 5.5 43.4% max, Claude Code Sep 2026 Official leaderboard
6 Kimi K3.1 Pro 43.4% max, Kimi Code Oct 2026 Official leaderboard
7 Gemini 4 Hydrogen 43.4% xhigh, Antigravity Cli Nov 2026 Official leaderboard
8 GPT-6 Astra 42.6% max, Codex Sep 2026 Official leaderboard
9 GPT-6.1 Terra 41.8% max, Codex Oct 2026 Official leaderboard
10 GPT-6.2 Sol 41.8% xhigh, Codex Nov 2026 Official leaderboard
11 Tencent Hy4 41% high, Claude Code Nov 2026 Official leaderboard
12 Gemini 4 Argon 40.2% high, ALE-Claw Sep 2026 Official leaderboard
13 Muse Wonder 38.5% high, Muse Code Oct 2026 Official leaderboard
14 DeepSeek V4.1 Flash 1017 36.1% high, OpenClaw Oct 2026 Official leaderboard
15 Kimi K3 35.3% max, Kimi Code Jul 2026 Official leaderboard
16 GPT-5.6 Sol 30.3% max, Codex Jul 2026 Official leaderboard
17 MiniMax-M3.1-Flash 29.5% max, Codex Oct 2026 Official leaderboard
18 GLM-5.3 26.2% max, ZCode Aug 2026 Official leaderboard
19 GPT-5.6 Luna 5.74% none, Codex Jul 2026 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6.2 Astra 47.5% xhigh, Codex Nov 2026 Official leaderboard
2 Claude Fable 5.2 46.9% max, Claude Code Nov 2026 Official leaderboard
3 Gemini 4 Carbon 45.9% high, ALE-Claw Oct 2026 Official leaderboard
4 Claude Opus 5.5 43.4% max, Claude Code Sep 2026 Official leaderboard
5 Kimi K3.1 Pro 43.4% max, Kimi Code Oct 2026 Official leaderboard
6 Gemini 4 Hydrogen 43.4% xhigh, Antigravity Cli Nov 2026 Official leaderboard
7 GPT-6 Astra 42.6% max, Codex Sep 2026 Official leaderboard
8 GPT-6.1 Terra 41.8% max, Codex Oct 2026 Official leaderboard
9 GPT-6.2 Sol 41.8% xhigh, Codex Nov 2026 Official leaderboard
10 Tencent Hy4 41% high, Claude Code Nov 2026 Official leaderboard
11 Gemini 4 Argon 40.2% high, ALE-Claw Sep 2026 Official leaderboard
12 Muse Wonder 38.5% high, Muse Code Oct 2026 Official leaderboard
13 DeepSeek V4.1 Flash 1017 36.1% high, OpenClaw Oct 2026 Official leaderboard
14 Kimi K3 35.3% max, Kimi Code Jul 2026 Official leaderboard
15 GPT-5.6 Sol 30.3% max, Codex Jul 2026 Official leaderboard
16 MiniMax-M3.1-Flash 29.5% max, Codex Oct 2026 Official leaderboard
17 GLM-5.3 26.2% max, ZCode Aug 2026 Official leaderboard
18 GPT-5.6 Luna 5.74% none, Codex Jul 2026 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6.2 Astra 47.5% xhigh, Codex Nov 2026 Official leaderboard
2 Claude Fable 5.2 46.9% max, Claude Code Nov 2026 Official leaderboard
3 Qwen4-Max 46.7% Nov 2026 Lab model card
4 Gemini 4 Carbon 45.9% high, ALE-Claw Oct 2026 Official leaderboard
5 Claude Opus 5.5 43.4% max, Claude Code Sep 2026 Official leaderboard
6 Kimi K3.1 Pro 43.4% max, Kimi Code Oct 2026 Official leaderboard
7 Gemini 4 Hydrogen 43.4% xhigh, Antigravity Cli Nov 2026 Official leaderboard
8 GPT-6 Astra 42.6% max, Codex Sep 2026 Official leaderboard
9 GPT-6.1 Terra 41.8% max, Codex Oct 2026 Official leaderboard
10 GPT-6.2 Sol 41.8% xhigh, Codex Nov 2026 Official leaderboard
11 Tencent Hy4 41% Nov 2026 Lab model card
12 Gemini 4 Argon 40.2% high, ALE-Claw Sep 2026 Official leaderboard
13 Muse Wonder 38.5% high, Muse Code Oct 2026 Official leaderboard
14 LongCat-2.5 38.5% Nov 2026 Lab model card
15 DeepSeek V4.1 Flash 1017 36.1% high, OpenClaw Oct 2026 Official leaderboard
16 Kimi K3 35.3% max, Kimi Code Jul 2026 Official leaderboard
17 GPT-5.6 Sol 30.3% max, Codex Jul 2026 Official leaderboard
18 MiniMax-M3.1-Flash 29.5% max, Codex Oct 2026 Official leaderboard
19 GLM-5.3 26.2% max, ZCode Aug 2026 Official leaderboard
20 GPT-5.6 Luna 5.74% none, Codex Jul 2026 Official leaderboard

Labs report scores for 3 models in their cards. Switch the scores above to rank them together.

Other ways to score it

The benchmark’s own leaderboard also reports Score, another way to score the whole test. Labs sometimes quote it instead of the headline. GPT-6.2 Astra leads it. Arrows show how far a model moved from its overall rank among the models we track.

Partial credit: the mean task score from 0 to 100, the board's Score column. Google and OpenAI quote it, and most other labs quote the pass rate. 18 models we track have a score here.

  1. 1 GPT-6.2 Astra 78.3%
  2. 2 Gemini 4 Carbon ↑1 77.4%
  3. 3 Kimi K3.1 Pro ↑2 74.4%
  4. 4 GPT-6 Astra ↑3 73.1%
  5. 5 GPT-6.2 Sol ↑4 73%
  6. 6 Claude Fable 5.2 ↓4 72.8%
  7. 7 Claude Opus 5.5 ↓3 71.6%
  8. 8 Gemini 4 Hydrogen ↓2 71.4%
  9. 9 GPT-6.1 Terra ↓1 70.4%
  10. 10 Tencent Hy4 70.3%

The best score, 47.5%, costs $3.05 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Cost frontier One model at each effort level
0% 10% 20% 30% 40% 50% $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6.2 Astra, xhigh GPT-6.1 Terra, max GPT-6.1 Terra, low GPT-5.6 Luna, none 0% 10% 20% 30% 40% 50% $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6.2 Astra, xhigh GPT-6.1 Terra, max

Source: Official Agents' Last Exam v1.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.2 Astra xhigh Codex 47.5% $3.05 20,922
Claude Fable 5.2 max Claude Code 46.9% $60 249,970
GPT-6.2 Astra max Codex 46.7% $4.68 40,888
Gemini 4 Carbon high ALE-Claw 45.9% $4.60 83,390
GPT-6.2 Astra high Codex 45.9% $2.17 14,336
Claude Opus 5.5 max Claude Code 43.4% $18 176,509
Kimi K3.1 Pro max Kimi Code 43.4% $7.00 69,747
Gemini 4 Hydrogen xhigh Antigravity Cli 43.4% $4.30 209,270
GPT-6 Astra max Codex 42.6% $8.99 41,823
GPT-6.1 Terra max Codex 41.8% $0.88 45,476
GPT-6.2 Sol xhigh Codex 41.8% $0.91 25,764
GPT-6.1 Terra xhigh Codex 41% $0.49 29,362
Tencent Hy4 high Claude Code 41% $0.68 160,318
GPT-6.2 Astra medium Codex 41% $1.14 6,969
GPT-6 Astra xhigh Codex 40.2% $7.56 31,652
GPT-6 Astra medium Codex 40.2% $6.21 21,331
Gemini 4 Argon high ALE-Claw 40.2% $4.66 53,659
GPT-6.1 Terra low Codex 40.2% $0.15 7,059
Gemini 4 Hydrogen high Antigravity Cli 40.2% $1.97 86,288
GPT-6.2 Sol max Codex 40.2% $1.39 39,568
GPT-6.2 Sol high Codex 40.2% $0.73 19,534
GPT-6 Astra high Codex 38.5% $6.87 26,879
GPT-6 Astra low Codex 38.5% $5.40 17,285
Muse Wonder high Muse Code 38.5% $3.12 38,829
GPT-6.2 Sol medium Codex 38.5% $0.61 13,827
GPT-6.2 Astra low Codex 37.7% $0.52 2,975
GPT-6.1 Terra none Codex 36.9% $0.32 10,128
DeepSeek V4.1 Flash 1017 high OpenClaw 36.1% $0.16 76,418
GPT-6.1 Terra high Codex 36.1% $0.29 16,140
Kimi K3 max Kimi Code 35.3% $4.76 61,795
DeepSeek V4.1 Flash 1017 max OpenClaw 35.3% $0.26 137,745
GPT-6.2 Sol low Codex 35.3% $0.29 6,153
GPT-6.1 Terra medium Codex 34.4% $0.23 11,233
Gemini 4 Hydrogen medium Antigravity Cli 33.6% $0.84 34,555
GPT-5.6 Sol max Codex 30.3% $3.33 26,004
MiniMax-M3.1-Flash max Codex 29.5% $0.49 142,504
MiniMax-M3.1-Flash max Claude Code 29.5% $1.12 148,463
GLM-5.3 max ZCode 26.2% $4.76 142,878
Gemini 4 Hydrogen low Antigravity Cli 23.8% $0.36 12,446
GPT-5.6 Luna none Codex 5.74% $0.12 8,723

The best score, 47.5%, costs $3.05 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Cost frontier One model at each effort level
0% 10% 20% 30% 40% 50% $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6.2 Astra, xhigh GPT-6.1 Terra, max GPT-6.1 Terra, low GPT-5.6 Luna, none 0% 10% 20% 30% 40% 50% $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6.2 Astra, xhigh GPT-6.1 Terra, max

Source: Official Agents' Last Exam v1.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.2 Astra xhigh Codex 47.5% $3.05 20,922
Claude Fable 5.2 max Claude Code 46.9% $60 249,970
GPT-6.2 Astra max Codex 46.7% $4.68 40,888
Gemini 4 Carbon high ALE-Claw 45.9% $4.60 83,390
GPT-6.2 Astra high Codex 45.9% $2.17 14,336
Claude Opus 5.5 max Claude Code 43.4% $18 176,509
Kimi K3.1 Pro max Kimi Code 43.4% $7.00 69,747
Gemini 4 Hydrogen xhigh Antigravity Cli 43.4% $4.30 209,270
GPT-6 Astra max Codex 42.6% $8.99 41,823
GPT-6.1 Terra max Codex 41.8% $0.88 45,476
GPT-6.2 Sol xhigh Codex 41.8% $0.91 25,764
GPT-6.1 Terra xhigh Codex 41% $0.49 29,362
Tencent Hy4 high Claude Code 41% $0.68 160,318
GPT-6.2 Astra medium Codex 41% $1.14 6,969
GPT-6 Astra xhigh Codex 40.2% $7.56 31,652
GPT-6 Astra medium Codex 40.2% $6.21 21,331
Gemini 4 Argon high ALE-Claw 40.2% $4.66 53,659
GPT-6.1 Terra low Codex 40.2% $0.15 7,059
Gemini 4 Hydrogen high Antigravity Cli 40.2% $1.97 86,288
GPT-6.2 Sol max Codex 40.2% $1.39 39,568
GPT-6.2 Sol high Codex 40.2% $0.73 19,534
GPT-6 Astra high Codex 38.5% $6.87 26,879
GPT-6 Astra low Codex 38.5% $5.40 17,285
Muse Wonder high Muse Code 38.5% $3.12 38,829
GPT-6.2 Sol medium Codex 38.5% $0.61 13,827
GPT-6.2 Astra low Codex 37.7% $0.52 2,975
GPT-6.1 Terra none Codex 36.9% $0.32 10,128
DeepSeek V4.1 Flash 1017 high OpenClaw 36.1% $0.16 76,418
GPT-6.1 Terra high Codex 36.1% $0.29 16,140
Kimi K3 max Kimi Code 35.3% $4.76 61,795
DeepSeek V4.1 Flash 1017 max OpenClaw 35.3% $0.26 137,745
GPT-6.2 Sol low Codex 35.3% $0.29 6,153
GPT-6.1 Terra medium Codex 34.4% $0.23 11,233
Gemini 4 Hydrogen medium Antigravity Cli 33.6% $0.84 34,555
GPT-5.6 Sol max Codex 30.3% $3.33 26,004
MiniMax-M3.1-Flash max Codex 29.5% $0.49 142,504
MiniMax-M3.1-Flash max Claude Code 29.5% $1.12 148,463
GLM-5.3 max ZCode 26.2% $4.76 142,878
Gemini 4 Hydrogen low Antigravity Cli 23.8% $0.36 12,446
GPT-5.6 Luna none Codex 5.74% $0.12 8,723

4 models have held number one on the official leaderboard since its release in October 2026 4 models have held number one with claims included since its release in October 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Anthropic Google DeepMind OpenAI Meituan Tencent Qwen
Own leaderboard 4 leaders
Lab claims 3 leaders
Qwen4-Max 46.7%

Sources: Official Agents' Last Exam v1.1 leaderboard and labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6.2 Astra Claude Opus 5.5 Kimi K3 GPT-5.6 Sol 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6.2 Astra Kimi K3

Sources: Official Agents' Last Exam v1.1 leaderboard and labs’ model cards

5 labs put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Google DeepMind
Meituan
Tencent
Qwen
OpenAI
DateLaunch postScoreAs the post writes it
Nov 18, 2026 Gemini 4 Neon 75.8 Agents' Last Exam v1.1
Nov 17, 2026 GPT-6.2 Sol 73 Agents' Last Exam v1.1 (GPT-6.2 Sol Max)
Nov 17, 2026 GPT-6.2 Astra 78.3 Agents' Last Exam v1.1 (GPT-6.2 Astra Xhigh, 77.6 at Max)
Nov 15, 2026 Qwen4-Max 46.7 Long-horizon valuable tasks, Agents' Last Exam v1.1, cited from its leaderboard
Nov 12, 2026 Gemini 4 Hydrogen 71.6 Agents' Last Exam v1.1 (Score)
Nov 4, 2026 Tencent Hy4 41 Agents' Last Exam v1.1*
Nov 2, 2026 LongCat-2.5 38.5 ALE (v1.1)
Oct 28, 2026 Gemini 4 Carbon 77.4 Agents' Last Exam v1.1 (Score)

2 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Nov 18, 2026 Gemini 4 Neon model card 75.8 Agents' Last Exam v1.1 (results table, p.4) Capabilities
Nov 10, 2026 Gemini 4 Carbon model card 77.4 Agents' Last Exam v1.1 (Score) (results table) Capabilities

5 lab claims next to the leaderboard

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 4 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Gemini 4 Hydrogen Score 71.6% Agents' Last Exam v1.1 (Score) 71.4% Leaderboard, xhigh, Antigravity Cli +0.2 pts
GPT-6.2 Sol Score 73% Agents' Last Exam v1.1 (GPT-6.2 Sol Max) 73% Leaderboard, max, Codex Same
GPT-6.2 Astra Score 78.3% Agents' Last Exam v1.1 (GPT-6.2 Astra Xhigh, 77.6 at Max) 78.3% Leaderboard, xhigh, Codex Same
Tencent Hy4 41% Agents' Last Exam v1.1* 41% Leaderboard, high, Claude Code Same
Gemini 4 Carbon Score 77.4% Agents' Last Exam v1.1 (Score) 77.4% Leaderboard, high, ALE-Claw Same

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Anthropic OpenAI Google DeepMind
Agents' Last Exam v1.1 9 leaders
Agents' Last Exam Replaced by v1.1, Oct 2026

Source: Official leaderboards

More in Economic work

All 210 in Economic work
  • JobBench In use 130 tasks from 35 occupations workers most want delegated. Agent works through messy reference files, graded by fact-anchored binary rubrics. GLM-5.3 61.4%
  • GDPval In use Real work tasks from 44 occupations with reference files. Experts blindly compare model deliverables with expert work. Metric is win-or-tie rate. Seed2.1 Pro 87.9%
  • GDPval-AA v2.1 In use Agents with shell and web access complete 220 GDPval gold tasks from 44 occupations. Blind pairwise comparisons give Elo ratings. Claude Opus 5.5 1,846
  • AA-Briefcase v1.1 In use Version 1.1 of Artificial Analysis's multi-hour office work evaluation, with Elo ratings fitted by Crowd-BT. Claude Opus 5.5 1,822
  • CoWorkBench In use Agent completes long-horizon productivity tasks across computer science, finance, law and medicine. Qwen in-house scoring (method unpublished). Claude Fable 5 75.9%
  • Vending-Bench 2 In use Agent runs a simulated vending-machine business for one year starting with $500. Score is the final bank balance, averaged over runs. Gemini 4 Neon $16,843

Data audit

All 12 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (19 entries), launch posts (8 scores) and system cards (2 scores).
  • Pass Models scored 20 models we track, plus 1 other entry on its leaderboard.
  • Pass Newest score The newest is from November 2026, for Gemini 4 Neon.
  • Pass Frontier models 7 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Pass Headroom The best is 47.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on agents-last-exam.org: 40 rows for models we track and 1 other entry at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 2 pass
  • Pass Lab claims All 5 pairs agree within 10 points.
  • Pass Known issues No published audit found problems.

How we check