DAYJOB
We show this table for reference; we do not rank on it.
DAYJOB is Surge AI's test of long professional assignments: 80 finance tasks and 50 healthcare tasks that an agent must finish to an expert checklist. This is the combined 130-task board on Kaggle.
Evaluation results published by Surge AI. How Kaggle Benchmarks reruns tests.
All-pass rate (%) on DAYJOB v1 — Latest evaluation November 4, 2026
We mirror the published all-pass rate (%) view for DAYJOB v1. Claude Opus 5.5 has the highest reported value at 24.5%. These are published source configurations. Compare the disclosed settings before treating a score difference as a model difference. We do not use these results to rank models overall.
Claude Opus 5.5
Max Reasoning; Evaluated 2026-11-03
Claude Sonnet 5.5
Max Reasoning; Evaluated 2026-11-03
Gemini 4 Carbon
High Reasoning; Evaluated 2026-11-03
19 source rowsAgenticCurrentDisplay onlyUpdated Latest evaluation November 4, 2026
All-pass rate (%) table (19 source rows)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Opus 5.5Creator not reportedMax Reasoning; Evaluated 2026-11-03 | 24.5% | Not reported | Unknown |
| 2 | Claude Sonnet 5.5Creator not reportedMax Reasoning; Evaluated 2026-11-03 | 21.7% | Not reported | Unknown |
| 3 | Gemini 4 CarbonCreator not reportedHigh Reasoning; Evaluated 2026-11-03 | 19.7% | Not reported | Unknown |
| 4 | GPT-6 AstraCreator not reportedMax Reasoning; Evaluated 2026-11-04 | 17.4% | Not reported | Unknown |
| 5 | Gemini 4 ArgonCreator not reportedHigh Reasoning; Evaluated 2026-11-03 | 15.7% | Not reported | Unknown |
| 6 | GPT-6.1 SolCreator not reportedMax Reasoning; Evaluated 2026-11-03 | 15.5% | Not reported | Unknown |
| 7 | Claude Opus 5Creator not reportedMax Reasoning; Evaluated 2026-11-03 | 11.7% | Not reported | Unknown |
| 8 | GPT-6 SolCreator not reportedMax Reasoning; Evaluated 2026-11-03 | 6.8% | Not reported | Unknown |
| 9 | GPT-5.6 SolCreator not reportedMax Reasoning; Evaluated 2026-11-03 | 4.2% | Not reported | Unknown |
| 10 | GPT-5.6 TerraCreator not reportedXHigh Reasoning; Evaluated 2026-11-03 | 2.8% | Not reported | Unknown |
| 11 | Gemini 3.8 FlashCreator not reportedHigh Reasoning; Evaluated 2026-11-03 | 2.6% | Not reported | Unknown |
| 12 | Claude Opus 4.8Creator not reportedMax Reasoning; Evaluated 2026-11-03 | 2.0% | Not reported | Unknown |
| 13 | Claude Sonnet 5Creator not reportedMax; Evaluated 2026-11-03 | 1.7% | Not reported | Unknown |
| 14 | GPT-5.5Creator not reportedXhigh Reasoning; Evaluated 2026-11-03 | 1.4% | Not reported | Unknown |
| 15 | GPT-6 LunaCreator not reportedMax Reasoning; Evaluated 2026-11-03 | 1.2% | Not reported | Unknown |
| 16 | GPT-5.6 LunaCreator not reportedMax Reasoning; Evaluated 2026-11-03 | 1.1% | Not reported | Unknown |
| 17 | Gemini 3 Flash PreviewCreator not reportedGemini 3 Flash High Reasoning; Evaluated 2026-11-03 | 0.0% | Not reported | Unknown |
| 18 | Gemini 3.1 Pro PreviewCreator not reportedHigh Reasoning; Evaluated 2026-11-03 | 0.0% | Not reported | Unknown |
| 19 | Gemini 3.5 Flash-LiteCreator not reportedHigh Reasoning; Evaluated 2026-11-03 | 0.0% | Not reported | Unknown |
How to read this leaderboard
Use it to judge whether a model can complete a whole piece of knowledge work unattended, not just answer a question about it. Compare the healthcare and finance pages when only one field matters to you.
Operator receipt: 19 sourced rows are currently displayable on this page; the highest published score among these 19 source rows is Claude Opus 5.5 at 24.5%, which does not establish a market leader.
Honest limit: Scores are low by design: models satisfy most individual criteria but rarely all of them. Results depend on the Kaggle agent harness and the judge model, and Surge AI keeps a separate leaderboard of its own. Kaggle reports close agreement between the two, not identity.
How to read DAYJOB
Each task is a short workplace request backed by a multi-file workspace. An agent (OpenHands SDK harness on Kaggle) must produce finished deliverables, which a judge model grades against an expert-written checklist of binary criteria. A trial passes only when every criterion passes, and each model gets five attempts per task; the score is the share of passing trials. The combined suite has 130 tasks across 11 subfields and 8,036 rubric criteria. The healthcare and finance tracks each have their own page.
Scores are low by design: models satisfy most individual criteria but rarely all of them. Results depend on the Kaggle agent harness and the judge model, and Surge AI keeps a separate leaderboard of its own. Kaggle reports close agreement between the two, not identity.
Snapshot
These are published source configurations. Compare the disclosed settings before treating a score difference as a model difference. The table retains each source configuration and its reported value.
19 source rows are shown for DAYJOB. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. DAYJOB is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About DAYJOB
Year
2026
Tasks
130 long-horizon tasks: 80 finance and 50 healthcare
Format
Agent deliverables graded all-or-nothing against expert rubrics
Difficulty
Long-horizon professional work
Each task is a short workplace request backed by a multi-file workspace. An agent (OpenHands SDK harness on Kaggle) must produce finished deliverables, which a judge model grades against an expert-written checklist of binary criteria. A trial passes only when every criterion passes, and each model gets five attempts per task; the score is the share of passing trials. The combined suite has 130 tasks across 11 subfields and 8,036 rubric criteria. The healthcare and finance tracks each have their own page. These results are display-only and do not enter overall or category rankings.
Freshness and provenance
Version
DAYJOB 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does DAYJOB measure?
DAYJOB is Surge AI's test of long professional assignments: 80 finance tasks and 50 healthcare tasks that an agent must finish to an expert checklist. This is the combined 130-task board on Kaggle.
Which source row has the highest reported DAYJOB value?
Claude Opus 5.5 has the highest reported DAYJOB value at 24.5%. These are published source configurations. Compare the disclosed settings before treating a score difference as a model difference. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on DAYJOB?
The Latest evaluation November 4, 2026 snapshot contains 19 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.