Benchmark Almanac

cognition-golden

Devin production (model unspecified) 74.2% tops its leaderboard o1-preview 51.8% leads the models we track
Not in model cards

Cognition tests coding agents on tasks modeled on real Devin use cases, in authentic development environments with fully autonomous grading. A train split serves self-improvement and a test split measures capability.

Cognition called it its primary internal benchmark when it compared o1-preview and GPT-4o inside Devin in 2024-09. The task set is not public.

Made by
Cognition
Released
September 2024
Kind
Building whole projects
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
o1-mini #140 on the capability index
Last model added
September 2024
Scores refreshed
November 12th, 2026
4
Entries on its own leaderboard, 2 models we track

Devin production (model unspecified) tops the leaderboard with 74.2%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 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. The best model we track is o1-preview, at 51.8%.

  1. 74.2 Devin production (model unspecified)
  2. 51.8 o1-preview
  3. 34.6 o1-mini
  4. 25.9 GPT-4o (snapshot unspecified)
  1. 1 Devin production (model unspecified) 74.2%
  2. 2 o1-preview 51.8%
  3. 3 o1-mini 34.6%
  4. 4 GPT-4o (snapshot unspecified) 25.9%

Source: Official cognition-golden leaderboard

#EntryScoreSettingDateSource
1 Devin production (model unspecified) 74.2% Official leaderboard
2 o1-preview 51.8% Devin-Base Sep 2024 Official leaderboard
3 o1-mini 34.6% Devin-Base Sep 2024 Official leaderboard
4 GPT-4o (snapshot unspecified) 25.9% Devin-Base Official leaderboard

Scores over time

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

On the benchmark’s own leaderboard
0% 20% 40% 60% 80% 100% Near the ceiling (85%) o1-preview 0% 20% 40% 60% 80% 100% Near the ceiling (85%) o1-preview

Source: Official cognition-golden leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Coding

All 282 in Coding
  • SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
  • LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests. DeepSeek-V4-Pro 93.5%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
  • DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
  • SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases. Seed2.1 Pro 59.8%
  • Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471