OpenHands Index
We show this table for reference; we do not rank on it.
A holistic coding-agent benchmark that evaluates AI agents across issue resolution, frontend work, greenfield development, testing, and information gathering.
Average agent score on OpenHands Index — September 28, 2026 snapshot
We mirror the published average agent score view for OpenHands Index. Claude Fable 5 leads the public snapshot at 81.0%, followed by Claude Opus 4.8 (71.9%) and Claude Opus 4.7 (Adaptive) (69.7%). We do not use these results to rank models overall.
Claude Fable 5
Anthropic
Claude Opus 4.8
Anthropic
Claude Opus 4.7 (Adaptive)
Anthropic
33 modelsAgenticCurrentDisplay onlyUpdated September 28, 2026 snapshot
Average agent score table (33 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Fable 5Anthropic | 81.0% | Not reported | Closed |
| 2 | Claude Opus 4.8Anthropic | 71.9% | Not reported | Closed |
| 3 | Claude Opus 4.7 (Adaptive)Anthropic | 69.7% | Not reported | Closed |
| 4 | Claude Opus 4.6Anthropic | 66.7% | Not reported | Closed |
| 5 | GPT-5.5OpenAI | 65.9% | Not reported | Closed |
| 6 | GPT-5.4OpenAI | 64.3% | Not reported | Closed |
| 7 | Gemini 3.5 FlashGoogle | 62.6% | Not reported | Closed |
| 8 | Claude Opus 4.5Anthropic | 60.6% | Not reported | Closed |
| 9 | Gemini 3.1 ProGoogle | 60.6% | Not reported | Closed |
| 10 | GPT-5.2OpenAI | 58.8% | Not reported | Closed |
| 11 | GPT-5.2-CodexOpenAI | 58.3% | Not reported | Closed |
| 12 | GLM-5.1Z.AI | 58.2% | Not reported | Open |
| 13 | MiniMax M3MiniMax | 57.2% | Not reported | Open |
| 14 | Kimi K2.6Moonshot AI | 57.1% | Not reported | Open |
| 15 | Claude Sonnet 4.5Anthropic | 53.0% | Not reported | Closed |
| 16 | Qwen3.6 PlusAlibaba | 52.9% | Not reported | Closed |
| 17 | GLM-5Z.AI | 49.4% | Not reported | Open |
| 18 | Kimi K2.5Moonshot AI | 49.2% | Not reported | Open |
| 19 | Gemini 3 ProGoogle | 49.0% | Not reported | Closed |
| 20 | Gemini 3 FlashGoogle | 49.0% | Not reported | Closed |
| 21 | DeepSeek V3.2 (Thinking)DeepSeek | 45.7% | Not reported | Open |
| 22 | MiniMax M2.5MiniMax | 45.2% | Not reported | Closed |
| 23 | Claude Sonnet 4.6Anthropic | 44.5% | Not reported | Closed |
| 24 | Qwen3 Coder NextAlibaba | 43.8% | Not reported | Unknown |
| 25 | MiniMax M2.7MiniMax | 43.4% | Not reported | Open |
| 26 | GLM-4.7Z.AI | 42.3% | Not reported | Open |
| 27 | MiniMax M2.1MiniMax | 41.2% | Not reported | Unknown |
| 28 | Kimi K2.5 (Reasoning)Moonshot AI | 41.0% | Not reported | Closed |
| 29 | DeepSeek V4 Pro 0813DeepSeek | 40.7% | Not reported | Open |
| 30 | Qwen3.5 FlashAlibaba | 38.1% | Not reported | Closed |
| 31 | Nemotron 3 Super 120B A12BNVIDIA | 36.2% | Not reported | Open |
| 32 | Trinity-Large-ThinkingArcee AI | 32.1% | Not reported | Open |
| 33 | Qwen3-Coder 480B-A35BAlibaba | 30.9% | Not reported | Open |
How OpenHands Index is shown here
BenchLM mirrors the official OpenHands Index REST API snapshot from September 28, 2026 snapshot. The source evaluates 33 coding-agent model variants across 5 categories: Issue Resolution, Frontend, Greenfield, Testing, Information Gathering.
OpenHands Index is display only on BenchLM. It is a valuable agentic software-engineering reference, but its rows combine model, SDK version, agent harness, cost, runtime, and per-benchmark result links, so BenchLM keeps it separate from weighted model-only rankings.
Snapshot
The published OpenHands Index snapshot places Claude Fable 5 first at 81.0%. The third row is 11.3 points behind. The broader top-10 range is 22.2 points, so the table still separates the published systems.
33 models have been evaluated on OpenHands Index. The benchmark falls in the Agentic category. OpenHands Index is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About OpenHands Index
Year
2025
Tasks
SWE-bench Verified, SWE-bench Multimodal, Commit0, SWT-bench Verified, and GAIA
Format
Macro-average across five coding-agent categories
Difficulty
Real-world software engineering agent tasks
BenchLM mirrors the official OpenHands Index REST API as a display-only agentic software-engineering benchmark. The source reports average agent score, cost, runtime, per-category scores, logs, and visualizations for each model and SDK version.
Freshness and provenance
Version
OpenHands Index 2025
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 OpenHands Index measure?
A holistic coding-agent benchmark that evaluates AI agents across issue resolution, frontend work, greenfield development, testing, and information gathering.
Which model leads the published OpenHands Index snapshot?
Claude Fable 5 currently leads the published OpenHands Index snapshot with 81.0% average agent score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on OpenHands Index?
The September 28, 2026 snapshot snapshot contains 33 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 5,600+ readers.
One email each week. Unsubscribe anytime.