Hugging Face's logo

Datasets:
OpenHands
/
openhands-index

Modalities:
Tabular
Text
Formats:
parquet
Size:
10K - 100K
Tags:
leaderboard
code
agents
benchmark
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
openhands-index / README.md
juan-all-hands's picture
juan-all-hands
Update leaderboard v2026.06.18-dea8a76 (34 models, 40643 instance rows)
625c370 verified
preview code
|
Raw
History Blame Contribute Delete
5.23 kB
metadata 
license: apache-2.0
pretty_name: OpenHands Index Leaderboard
tags:
  - leaderboard
  - code
  - agents
  - benchmark
dataset_info:
  - config_name: default
    version: 2026.6.18
    description: >-
      Snapshot of the OpenHands Index leaderboard built from
      openhands-index-results commit dea8a76b5578e2271ee35d4bc8ddfc0eecb00edf on
      2026-06-18 23:39:05 UTC.
  - config_name: instances
    version: 2026.6.18
    description: >-
      Per-instance benchmark outcomes (resolved/cost) for every model in the
      `default` config, sourced from `results/<model>/instance_results/*.json`
      sidecars at commit dea8a76b5578e2271ee35d4bc8ddfc0eecb00edf.
configs:
  - config_name: default
    data_files:
      - split: test
        path: test.parquet
  - config_name: instances
    data_files:
      - split: test
        path: instances.parquet

OpenHands Index — Leaderboard Snapshot

Auto-published from https://github.com/OpenHands/openhands-index-results on every push to main. Matches the table shown at the OpenHands Index Space.

from datasets import load_dataset

# leaderboard (one row per model)
ds = load_dataset("OpenHands/openhands-index", split="test")
ds.info.version          # → "2026.06.18-dea8a76"

# per-instance outcomes (one row per model × instance)
instances = load_dataset("OpenHands/openhands-index", "instances", split="test")

# pin to this exact snapshot
ds = load_dataset("OpenHands/openhands-index", split="test", revision="v2026.06.18-dea8a76")

Categories

Category Backing benchmark
Issue Resolution SWE-Bench
Frontend SWE-Bench Multimodal
Greenfield Commit0
Testing SWT-Bench
Information Gathering GAIA

average_score is the mean of the per-benchmark scores actually completed. categories_completed tells you how many benchmarks the model has run.

Configs

default — leaderboard (one row per model)

The aggregate table backing the leaderboard Space. Use this for ranking, averages, and per-category scores.

instances — per-instance outcomes (one row per model × benchmark instance)

Long-form table of every benchmark instance's outcome for every model in default. Join to default on id.

Column Type Notes
id string Matches default.id (e.g. OpenHands/GPT-5.5)
agent_name string Display name from metadata.json
agent_type string Currently always OpenHands (see #1145)
language_model string LLM identifier
benchmark string One of swe-bench, swe-bench-multimodal, commit0, swt-bench, gaia
category string Human-facing category label
instance_id string Benchmark-specific instance ID
resolved bool? true / false / null when the archive didn't record an outcome
cost float? USD; null when unavailable

This snapshot

Top 15 by average score

language_model sdk_version agent_name average_score categories_completed release_date
claude-fable-5 v1.18.1 OpenHands 81.00 5 2026-06-09
claude-opus-4-8 v1.18.1 OpenHands 71.88 5 2026-05-28
claude-opus-4-7 v1.11.0 OpenHands 69.66 5 2026-04-16
claude-opus-4-6 v1.11.0 OpenHands 66.72 5 2026-02-05
GPT-5.5 v1.18.1 OpenHands 65.94 5 2026-04-23
GPT-5.4 v1.13.0 OpenHands 64.28 5 2026-03-05
Gemini-3.5-Flash v1.24.0 OpenHands 62.64 5 2026-05-19
claude-opus-4-5 v1.8.3 OpenHands 60.58 5 2025-11-24
Gemini-3.1-Pro v1.11.5 OpenHands 60.56 5 2026-01-15
GPT-5.2 v1.8.3 OpenHands 58.84 5 2025-12-11
GPT-5.2-Codex v1.8.3 OpenHands 58.28 5 2025-12-18
GLM-5.1 v1.11.5 OpenHands 58.24 5 2026-04-07
MiniMax-M3 v1.18.1 OpenHands 57.20 5 2026-06-01
Kimi-K2.6 v1.11.5 OpenHands 57.14 5 2026-04-20
claude-sonnet-4-5 v1.8.3 OpenHands 53.00 5 2025-09-29