Mercury 2.5: benchmark results
Inception's Mercury 2.5 diffusion LLM: about 1,100 tokens/s with tunable reasoning and parallel tool calls. Served only through Inception's API, Baseten and OpenRouter; no published weights. Provider: Inception. Released 2026-09-08. Access: API.
Aggregate Rating 1599 ± 31, rank #452 of 1635 rated models, from 66 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA Long Context Reasoning | 71.67 | Accuracy (%) | 64.1 |
| Wolfram LLM Benchmarking Project | 50.7 | Correct Functionality (%) | 63.8 |
| Vals AI LegalBench | 83.06 | Accuracy (%) | 59.8 |
| ObviousBench | 94.44 | Answer pass³ (%) | 55.7 |
| AA Humanity's Last Exam | 11.77 | Accuracy (%) | 53.9 |
| AIIQ Composite IQ | 100 | Composite IQ (imported) | 27.7 |
| AA GDP.pdf - Insurance - Criterion Pass Rate | 53.86 | Criterion Pass Rate (%) | 24.5 |
| AA GDP.pdf - Legal - Criterion Pass Rate | 57.19 | Criterion Pass Rate (%) | 24.5 |
| AA GDP.pdf - HR - Criterion Pass Rate | 55.91 | Criterion Pass Rate (%) | 24.1 |
| AA GDPval | 481.79 | ELO | 23.1 |
| Tinybird AI SQL Benchmark - Exactness | 41.89 | Result exactness vs human reference queries (0-100) | 22.9 |
| AutomationBench-AA - Operations - Raw Objective Completion | 8.25 | Raw Objective Completion (%) | 22.3 |
Interactive version: theaggregate.ai/model?slug=mercury-2-5 · How It Works · Data refreshed daily, snapshot 2026-11-05.