Benchmark Almanac

Numerical results extraction from randomized controlled trials

DeepSeek V4.1 Flash 77.9% leads since September 2026
Not in model cards

Models extract group sizes, event counts, means, and standard deviations from trial reports. The test contains 656 treatment comparisons across 110 papers.

67AI Lab tested seven models on 648 comparisons from 109 available papers. Its table reports the share of correctly filled cells.

Released
May 2024
Kind
Medicine
Level
University
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Measures
Capability
Language
English
In and out
Documents → Tables
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
Strongest model tested
Claude Opus 5.5 #5 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026
7
Models scored on 67AI Lab’s leaderboard

DeepSeek V4.1 Flash leads with 77.9% among models released by September 2026

Best score per model from 67AI Lab’s leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

The newest model with a score here came out in September 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 DeepSeek V4.1 Flash Exact (median) 77.9%
  2. 2 Claude Opus 5.5 Exact (median) 75.8%
  3. 3 Qwen 3.6 35B-A3B Greedy 75.5%
  4. 4 Qwen3-Coder-Next Greedy 74.7%
  5. 5 Gemma 4 26B A4B Greedy 73.8%
  6. 6 Qwen 3.8 27B Greedy 72.3%
  7. 7 Nemotron 3.5 Lightning 30B-A3B Greedy 62.7%

Source: 67AI Lab’s leaderboard

#ModelScoreSettingDateSource
1 DeepSeek V4.1 Flash 77.9% Exact (median) Sep 2026 Official leaderboard
2 Claude Opus 5.5 75.8% Exact (median) Sep 2026 Official leaderboard
3 Qwen 3.6 35B-A3B 75.5% Greedy Apr 2026 Official leaderboard
4 Qwen3-Coder-Next 74.7% Greedy Feb 2026 Official leaderboard
5 Gemma 4 26B A4B 73.8% Greedy Apr 2026 Official leaderboard
6 Qwen 3.8 27B 72.3% Greedy Aug 2026 Official leaderboard
7 Nemotron 3.5 Lightning 30B-A3B 62.7% Greedy Aug 2026 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Qwen 3.6 35B-A3B
Exact (medi…
75.5%
Qwen3-Coder-Next
Exact (medi…
74.7%
Gemma 4 26B A4B
Exact (medi…
73.8%
Qwen 3.8 27B
Exact (medi…
72.3%
Nemotron 3.5 Lightning 30B-A3B
Exact (medi… Greedy
62.7%

Source: 67AI Lab’s leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
DeepSeek V4.1 Flash Exact (median) 77.9%
Claude Opus 5.5 Exact (median) 75.8%
Qwen 3.6 35B-A3B Greedy 75.5%
Qwen3-Coder-Next Greedy 74.7%
Qwen 3.6 35B-A3B Exact (median) 74.2%
Gemma 4 26B A4B Greedy 73.8%
Qwen3-Coder-Next Exact (median) 73.3%
Gemma 4 26B A4B Exact (median) 72.9%
Qwen 3.8 27B Greedy 72.3%
Qwen 3.8 27B Exact (median) 71.5%
Nemotron 3.5 Lightning 30B-A3B Greedy 62.7%
Nemotron 3.5 Lightning 30B-A3B Exact (median) 59.6%

3 models have held number one on 67AI Lab’s leaderboard since its release in May 2024

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

Qwen DeepSeek
67AI Lab leaderboard Not tested since Sep 2026

Source: 67AI Lab’s 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 67AI Lab’s leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) DeepSeek V4.1 Flash Qwen 3.6 35B-A3B Qwen3-Coder-Next 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) DeepSeek V4.1 Flash Qwen3-Coder-Next

Source: 67AI Lab’s 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 Law, finance and medicine

All 637 in Law, finance and medicine
  • MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy. Qwen 3.8 Max 80.4%
  • HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
  • HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
  • FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy. Ling-3.1-flash 81.7%
  • KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. o1-preview 92.7%
  • Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%

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 (7 entries).
  • Pass Models scored 7 models we track.
  • Pass Newest score The newest is from September 2026, for Claude Opus 5.5.
  • Note Frontier models Only 1 of the top 10 has a score. The strongest tested is Claude Opus 5.5 (#5).
  • Pass Headroom The best is 77.9%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on 67AI Lab: 12 rows for models we track 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