Benchmark Almanac

Agent Initiative Benchmark

GPT-6 Astra 8.97 leads since September 2026
Not in model cards

Agents inspect workplace situations and decide whether to act without an explicit task. Two model judges score responses to 50 situations from 0 to 10.

Each run allows 180 seconds in a synthetic company with 11 services. Some situations require correctly taking no action. The release day remains unconfirmed.

Made by
Deepnote
Kind
Simulated companies
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
All public
Measures
Propensity
Language
English
In and out
Text → Actions, Text
Interaction
Agentic
Environment
APIs and tools
Items from
Generated
Refresh
One fixed set
Run by
Anyone
In views
Agents
Strongest model tested
Claude Fable 5.2 #2 on the capability index
Last model added
November 2026
Scores refreshed
November 18th, 2026
12
Models scored on its own leaderboard

GPT-6 Astra leads with 8.97

Best score per model 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.

Source: Official Agent Initiative Benchmark leaderboard

#ModelScoreSettingDateSource
1 GPT-6 Astra 8.97 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Sep 2026 Official leaderboard
2 GPT-6.1 Sol 8.65 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Sep 2026 Official leaderboard
3 Claude Fable 5.2 8.17 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Nov 2026 Official leaderboard
4 Claude Opus 5.5 7.84 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Sep 2026 Official leaderboard
5 Claude Fable 5.1 7.50 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Sep 2026 Official leaderboard
6 Claude Sonnet 5.5 6.65 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Sep 2026 Official leaderboard
7 GPT-6 Luna 6.38 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Sep 2026 Official leaderboard
8 DeepSeek V4.1 Flash 6.35 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Sep 2026 Official leaderboard
9 Mistral Large 4 6.09 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Oct 2026 Official leaderboard
10 Claude Haiku 5.5 6.05 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Oct 2026 Official leaderboard
11 Gemini 4 Argon 5.60 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Sep 2026 Official leaderboard
12 Grok 4.7 4.51 180-second limit; 50 situations; two judges, Deepnote Agent Initiative Benchmark Sep 2026 Official leaderboard

The best score, 8.97, costs $0.69 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.1 Sol at 180-second limit; 50 situations; two judges comes within 0.3 of the best for $0.11 a task, 6 times 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
0.00 2.00 4.00 6.00 8.00 10.0 $0.001 $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-6 Astra, 180-second limit; 50 situations; two judges GPT-6 Luna, 180-second limit; 50 situations; two judges 0.00 2.00 4.00 6.00 8.00 10.0 $0.001 $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-6 Astra, 180-second limit; 50 situations; two judges GPT-6 Luna, 180-second limit; 50 situations; two judges

Source: Official Agent Initiative Benchmark leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 8.97 $0.69
GPT-6.1 Sol 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 8.65 $0.11
Claude Fable 5.2 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 8.17 $1.87
Claude Opus 5.5 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 7.84 $1.28
Claude Fable 5.1 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 7.50 $3.02
Claude Sonnet 5.5 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 6.65 $0.42
GPT-6 Luna 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 6.38 $0.007
DeepSeek V4.1 Flash 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 6.35 $0.03
Mistral Large 4 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 6.09 $0.08
Claude Haiku 5.5 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 6.05 $0.07
Gemini 4 Argon 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 5.60 $0.40
Grok 4.7 180-second limit; 50 situations; two judges Deepnote Agent Initiative Benchmark 4.51 $0.42

2 models have held number one on the official leaderboard since 2023

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic OpenAI
Own leaderboard 2 leaders

Source: Official Agent Initiative Benchmark leaderboard

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
4.00 5.00 6.00 7.00 8.00 9.00 2027 4.00 5.00 6.00 7.00 8.00 9.00 2027

Source: Official Agent Initiative Benchmark leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

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
  • Agents' Last Exam v1.1 In use Agents complete 122 public long-horizon professional workflow tasks across 55 sub-industries. Deterministic graders check outcomes against hidden references. GPT-6.2 Astra 47.5%
  • 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%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (12 entries).
  • Pass Models scored 12 models we track.
  • Pass Newest score The newest is from November 2026, for Claude Fable 5.2.
  • Pass Frontier models 7 of the top 10 have a score, led by Claude Fable 5.2 (#2).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on deepnote.com: 12 rows for models we track at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check