Corp Speak Arena
We show this table for reference; we do not rank on it.
Kaggle's Corp Speak Arena has models rewrite 100 everyday sentences in corporate language, and people vote for the better rewrite.
Evaluation results published by Kaggle. How Kaggle Benchmarks reruns tests.
Arena rating on Corp Speak Arena v1 — Latest evaluation November 10, 2026
We mirror the published arena rating view for Corp Speak Arena v1. GLM-5.3 has the highest reported value at 1731. A rating gap smaller than the two rows' intervals is not a reliable difference. We do not use these results to rank models overall.
GLM-5.3
Max Reasoning; 15 votes; ±186; Evaluated 2026-11-10
Claude Opus 5
Max Reasoning; 9 votes; ±220; Evaluated 2026-11-10
Claude Opus 4.7
Max Reasoning; 8 votes; ±212; Evaluated 2026-11-10
24 source rowsCross-domainCurrentDisplay onlyUpdated Latest evaluation November 10, 2026
Arena rating table (24 source rows)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | GLM-5.3Creator not reportedMax Reasoning; 15 votes; ±186; Evaluated 2026-11-10 | 1731 | Not reported | Unknown |
| 2 | Claude Opus 5Creator not reportedMax Reasoning; 9 votes; ±220; Evaluated 2026-11-10 | 1696 | Not reported | Unknown |
| 3 | Claude Opus 4.7Creator not reportedMax Reasoning; 8 votes; ±212; Evaluated 2026-11-10 | 1644 | Not reported | Unknown |
| 4 | Claude Opus 5.5Creator not reportedMax Reasoning; 21 votes; ±157; Evaluated 2026-11-10 | 1616 | Not reported | Unknown |
| 5 | Gemini 3.1 Pro PreviewCreator not reportedHigh Reasoning; 14 votes; ±182; Evaluated 2026-11-10 | 1578 | Not reported | Unknown |
| 6 | Grok 4.7Creator not reportedXhigh Reasoning; 18 votes; ±166; Evaluated 2026-11-10 | 1565 | Not reported | Unknown |
| 7 | Claude Sonnet 5.5Creator not reportedMax Reasoning; 22 votes; ±158; Evaluated 2026-11-10 | 1562 | Not reported | Unknown |
| 8 | GPT-6.1 SolCreator not reportedMax Reasoning; 26 votes; ±145; Evaluated 2026-11-10 | 1548 | Not reported | Unknown |
| 9 | Gemini 3.8 FlashCreator not reportedHigh Reasoning; 7 votes; ±225; Evaluated 2026-11-10 | 1542 | Not reported | Unknown |
| 10 | Claude Fable 5.1Creator not reportedMax Reasoning; 15 votes; ±171; Evaluated 2026-11-10 | 1541 | Not reported | Unknown |
| 11 | GPT-6 AstraCreator not reportedMax Reasoning; 19 votes; ±161; Evaluated 2026-11-10 | 1540 | Not reported | Unknown |
| 12 | Gemini 4 ArgonCreator not reportedHigh Reasoning; 21 votes; ±156; Evaluated 2026-11-10 | 1498 | Not reported | Unknown |
| 13 | GPT-5.6 SolCreator not reportedMax Reasoning; 9 votes; ±202; Evaluated 2026-11-10 | 1481 | Not reported | Unknown |
| 14 | Claude Sonnet 5Creator not reportedHigh Reasoning; 10 votes; ±195; Evaluated 2026-11-10 | 1462 | Not reported | Unknown |
| 15 | GPT-5 ProCreator not reportedHigh Reasoning; 12 votes; ±185; Evaluated 2026-11-10 | 1460 | Not reported | Unknown |
| 16 | GPT-5.2Creator not reportedXhigh Reasoning; 8 votes; ±208; Evaluated 2026-11-10 | 1457 | Not reported | Unknown |
| 17 | GPT-5.6 TerraCreator not reportedHigh Reasoning; 18 votes; ±165; Evaluated 2026-11-10 | 1451 | Not reported | Unknown |
| 18 | GPT-5.5Creator not reported GPT-5.5 High Reasoning; 20 votes; ±163; Evaluated 2026-11-10 | 1441 | Not reported | Unknown |
| 19 | Claude Opus 4.7Creator not reportedHigh Reasoning; 6 votes; ±227; Evaluated 2026-11-10 | 1437 | Not reported | Unknown |
| 20 | GPT-5.6 TerraCreator not reportedGPT 5.6 Terra Max Reasoning; 10 votes; ±200; Evaluated 2026-11-10 | 1434 | Not reported | Unknown |
| 21 | Qwen 3.8 MaxCreator not reportedXhigh Reasoning; 13 votes; ±185; Evaluated 2026-11-10 | 1416 | Not reported | Unknown |
| 22 | GPT-5Creator not reportedHigh Reasoning; 6 votes; ±235; Evaluated 2026-11-10 | 1367 | Not reported | Unknown |
| 23 | GPT-5.4Creator not reportedXhigh Reasoning; 13 votes; ±196; Evaluated 2026-11-10 | 1364 | Not reported | Unknown |
| 24 | GPT-6 LunaCreator not reportedMax Reasoning; 15 votes; ±202; Evaluated 2026-11-10 | 1218 | Not reported | Unknown |
How to read this leaderboard
Read the rating as a preference ranking for rewrites in corporate language among the models on this board, with its vote count and interval. It does not measure accuracy.
Operator receipt: 24 sourced rows are currently displayable on this page; the highest published score among these 24 source rows is GLM-5.3 at 1731, which does not establish a market leader.
Honest limit: The rating moves as votes arrive, and Kaggle publishes no confidence level for its interval. Rows with few votes have wide intervals, so neighbouring models are often not separable. Voters are self-selected Kaggle visitors. The board declares no license; we keep published aggregates with attribution and no task or output contents.
How to read Corp Speak Arena
A response qualifies if it is a plain-text answer of 5 to 100 words. It is a test of tone and style transfer in business writing. Kaggle shows two anonymous outputs for the same prompt and a person picks the better one. The Arena column is the relative rating Kaggle fits to those votes, published with an interval; it is not a percentage.
The rating moves as votes arrive, and Kaggle publishes no confidence level for its interval. Rows with few votes have wide intervals, so neighbouring models are often not separable. Voters are self-selected Kaggle visitors. The board declares no license; we keep published aggregates with attribution and no task or output contents.
Snapshot
A rating gap smaller than the two rows' intervals is not a reliable difference. The table retains each source configuration and its reported value.
24 source rows are shown for Corp Speak Arena. The benchmark falls in the Cross-domain category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Corp Speak Arena is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Corp Speak Arena
Year
2026
Tasks
100 plain sentences in 10 categories, rated by human votes
Format
Relative rating fitted to human pairwise votes
Difficulty
Tone and style rewriting
A response qualifies if it is a plain-text answer of 5 to 100 words. It is a test of tone and style transfer in business writing. Kaggle shows two anonymous outputs for the same prompt and a person picks the better one. The Arena column is the relative rating Kaggle fits to those votes, published with an interval; it is not a percentage. These results are display-only and do not enter overall or category rankings.
Freshness and provenance
Version
Corp Speak Arena 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 Corp Speak Arena measure?
Kaggle's Corp Speak Arena has models rewrite 100 everyday sentences in corporate language, and people vote for the better rewrite.
Which source row has the highest reported Corp Speak Arena value?
GLM-5.3 has the highest reported Corp Speak Arena value at 1731. A rating gap smaller than the two rows' intervals is not a reliable difference. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Corp Speak Arena?
The Latest evaluation November 10, 2026 snapshot contains 24 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.