Benchmark Almanac

MANDOS Session 2

Claude Sonnet 4.6 96.9% leads since September 2026
Not in model cards
Version Session 2

Models make process safety decisions across four private industrial scenarios. A model judge assesses each run. The leaderboard reports the share of passing runs.

RelayForge scores seven models on 224 runs. Its separate four-model writeup covers 128 runs. The scenarios differ from the June incidents with matching identifiers.

Made by
RelayForge
Released
September 2026
Kind
Tasks from real jobs
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Measures
Capability
Language
English
In and out
Text → Text, Actions
Interaction
Agentic
Environment
A simulator
Items from
Another benchmark
Run by
A host
Occupations
Architecture and engineering, Production
Strongest model tested
Claude Sonnet 4.6 #57 on the capability index
Last model added
April 2026
Scores refreshed
November 20th, 2026
7
Entries on its own leaderboard, 6 models we track

Claude Sonnet 4.6 leads with 96.9% among models released by April 2026 Claude Sonnet 4.6 leads with 96.9% among models released by April 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. 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.

The newest model with a score here came out in April 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Claude Sonnet 4.6 96.9%
  2. 2 Kimi K2.6 93.8%
  3. 3 openrouter-deepseek-chat 90.6%
  4. 4 DeepSeek-R1 90.6%
  5. 5 Llama 3.3 70B 84.4%
  6. 6 GPT-4o Mar 2025 84.4%
  7. 7 GPT-4o mini 59.4%

Source: Official MANDOS Session 2 leaderboard

#EntryScoreSettingDateSource
1 Claude Sonnet 4.6 96.9% Feb 2026 Official leaderboard
2 Kimi K2.6 93.8% Apr 2026 Official leaderboard
3 openrouter-deepseek-chat 90.6% Official leaderboard
4 DeepSeek-R1 90.6% Jan 2025 Official leaderboard
5 Llama 3.3 70B 84.4% Dec 2024 Official leaderboard
6 GPT-4o Mar 2025 84.4% Mar 2025 Official leaderboard
7 GPT-4o mini 59.4% Jul 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Sonnet 4.6 96.9% Feb 2026 Official leaderboard
2 Kimi K2.6 93.8% Apr 2026 Official leaderboard
3 DeepSeek-R1 90.6% Jan 2025 Official leaderboard
4 Llama 3.3 70B 84.4% Dec 2024 Official leaderboard
5 GPT-4o Mar 2025 84.4% Mar 2025 Official leaderboard
6 GPT-4o mini 59.4% Jul 2024 Official leaderboard

Number one over time

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

Anthropic
Own leaderboard Passed 85% in Sep 2026, by Claude Sonnet 4.6

Source: Official MANDOS Session 2 leaderboard

DeepSeek-R1 on the official leaderboard passed 85% in January 2025

Models released before it came out already scored that high. 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 Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Claude Sonnet 4.6 DeepSeek-R1 Llama 3.3 70B GPT-4o mini 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Claude Sonnet 4.6 DeepSeek-R1

Source: Official MANDOS Session 2 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.

Versions

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 8 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access The tasks are private.
  • Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
  • Pass Independent runs Scores from the official leaderboard (7 entries).
  • Pass Models scored 6 models we track, plus 1 other entry on its leaderboard.
  • Note Newest score The newest came 7 months ago, in April 2026, for Kimi K2.6. That’s expected for a saturated test.
  • Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Claude Sonnet 4.6 (#57).
  • Note Headroom Saturated. The best is 96.9%, past 85% since January 2025.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 6 rows for models we track and 1 other entry at the last read.
  • Pass Last read November 20th, 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