ReactBench v1 (ReactBench)
We show this table for reference; we do not rank on it.
A coding-agent benchmark for realistic React work, with rubrics that check production concerns such as performance, accessibility, correctness, and code quality.
ReactBench score (Pass@1) on ReactBench — October 28, 2026 snapshot
We mirror the published reactbench score (pass@1) view for ReactBench. Claude Opus 5.5 leads the public snapshot at 54.4%, followed by Claude Opus 5.5 (50.8%) and Fable 5.1 · Max (49.2%). We do not use these results to rank models overall.
Claude Opus 5.5
Anthropic
Claude Opus 5.5
Anthropic
Fable 5.1 · Max
Anthropic
49 modelsCodingCurrentDisplay onlyUpdated October 28, 2026 snapshot
ReactBench score (Pass@1) table (49 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropic | 54.4% | Not reported | Closed |
| 2 | Claude Opus 5.5Anthropic | 50.8% | Not reported | Closed |
| 3 | Fable 5.1 · MaxAnthropic | 49.2% | Not reported | Unknown |
| 4 | GPT-5.6 SolOpenAI | 46.7% | Not reported | Closed |
| 5 | Fable 5.1 · XHighAnthropic | 46.2% | Not reported | Unknown |
| 6 | Claude Opus 5.5Anthropic | 46.2% | Not reported | Closed |
| 7 | Fable 5.1 · HighAnthropic | 44.1% | Not reported | Unknown |
| 8 | Claude Opus 5.5Anthropic | 43.6% | Not reported | Closed |
| 9 | Claude Opus 5Anthropic | 42.1% | Not reported | Closed |
| 10 | Claude Opus 5Anthropic | 41.5% | Not reported | Closed |
| 11 | GPT-5.6 SolOpenAI | 41.0% | Not reported | Closed |
| 12 | GPT-5.6 SolOpenAI | 36.9% | Not reported | Closed |
| 13 | Qwen 3.8 Max · XHighAlibaba | 36.4% | Not reported | Unknown |
| 14 | Claude Opus 5Anthropic | 36.4% | Not reported | Closed |
| 15 | Apex · MediumCallstack | 35.4% | Not reported | Unknown |
| 16 | DeepSeek V4.1 Flash 1017DeepSeek | 35.4% | Not reported | Open |
| 17 | GPT-5.6 SolOpenAI | 34.9% | Not reported | Closed |
| 18 | Apex · XHighCallstack | 34.4% | Not reported | Unknown |
| 19 | Claude Opus 5Anthropic | 34.4% | Not reported | Closed |
| 20 | Gemini 3.8 FlashGoogle | 34.4% | Not reported | Closed |
| 21 | Fable 5.1 · MediumAnthropic | 33.8% | Not reported | Unknown |
| 22 | Grok 4.6xAI | 33.8% | Not reported | Closed |
| 23 | Grok 4.6xAI | 33.8% | Not reported | Closed |
| 24 | DeepSeek V4 Pro 0813DeepSeek | 33.8% | Not reported | Open |
| 25 | GPT-5.6 LunaOpenAI | 33.8% | Not reported | Closed |
| 26 | GPT-5.6 SolOpenAI | 33.8% | Not reported | Closed |
| 27 | Kimi K3Moonshot AI | 33.3% | Not reported | Open |
| 28 | Fable 5.1 · LowAnthropic | 32.8% | Not reported | Unknown |
| 29 | GLM-5.3-FlashZ.AI | 32.8% | Not reported | Open |
| 30 | Claude Sonnet 5Anthropic | 32.3% | Not reported | Closed |
| 31 | Grok 4.7xAI | 32.3% | Not reported | Closed |
| 32 | GPT-5.6 LunaOpenAI | 32.3% | Not reported | Closed |
| 33 | Grok 4.7xAI | 31.8% | Not reported | Closed |
| 34 | Grok 4.7xAI | 31.8% | Not reported | Closed |
| 35 | Claude Sonnet 5Anthropic | 31.3% | Not reported | Closed |
| 36 | GPT-5.6 LunaOpenAI | 31.3% | Not reported | Closed |
| 37 | Claude Opus 5.5Anthropic | 30.8% | Not reported | Closed |
| 38 | GLM-5.3Z.AI | 30.8% | Not reported | Open |
| 39 | Claude Sonnet 5Anthropic | 30.3% | Not reported | Closed |
| 40 | Apex · LowCallstack | 29.7% | Not reported | Unknown |
| 41 | Grok 4.6xAI | 29.2% | Not reported | Closed |
| 42 | Grok 4.6xAI | 28.7% | Not reported | Closed |
| 43 | Grok 4.7xAI | 27.7% | Not reported | Closed |
| 44 | Claude Opus 5Anthropic | 27.2% | Not reported | Closed |
| 45 | Muse Spark 1.2Meta | 23.6% | Not reported | Closed |
| 46 | Claude Sonnet 5Anthropic | 22.6% | Not reported | Closed |
| 47 | GPT-5.6 LunaOpenAI | 20.0% | Not reported | Closed |
| 48 | Claude Sonnet 5Anthropic | 16.9% | Not reported | Closed |
| 49 | GPT-5.6 LunaOpenAI | 10.8% | Not reported | Closed |
How we show ReactBench
We mirror the official ReactBench table from Million: 49 effort variants across 39 React tasks. The source score is a weighted rubric aggregate, and solutions that miss a blocking criterion receive zero.
ReactBench tests production React work that ordinary behavior checks can miss, including performance, accessibility, and code quality. The rows also include an agent setup and reasoning-effort choice, so we preserve the published variants and costs without adding the scores to weighted coding rankings.
Snapshot
The published ReactBench snapshot places Claude Opus 5.5 first at 54.4%. The third row is 5.2 points behind. The broader top-10 range is 12.9 points, so the table still separates the published systems.
49 models have been evaluated on ReactBench. The benchmark falls in the Coding category. ReactBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About ReactBench
Year
2026
Tasks
51 production React tasks
Format
Pass@1 weighted rubric score
Difficulty
Production frontend engineering
ReactBench uses 51 tasks drawn from real React repositories. Its weighted rubric score assigns zero to solutions that miss a blocking criterion. We mirror each published reasoning-effort variant and its average rollout cost, but keep the results out of weighted coding scores because the agent setup and effort level are part of the row.
ReactBench contributes task-specific evidence. The frontend model shortlist puts that evidence beside the model choice without turning one score into an app-quality guarantee.
Freshness and provenance
Version
ReactBench 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 ReactBench measure?
A coding-agent benchmark for realistic React work, with rubrics that check production concerns such as performance, accessibility, correctness, and code quality.
Which model leads the published ReactBench snapshot?
Claude Opus 5.5 currently leads the published ReactBench snapshot with 54.4% reactbench score (pass@1). BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on ReactBench?
The October 28, 2026 snapshot snapshot contains 49 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 6,100+ subscribers.
One email each week. Unsubscribe anytime.