API test results
We run Veridrop, an open-source tool that checks AI APIs, on our own Cloudflare Workers and used it to test the four models on the RoamAI API. It checks whether the model behind an API is really Claude, whether features such as PDF input and tool use still work, and whether responses follow the official API format.
You can test any AI API with the same checks yourself: veridrop.roamai.cloud ↗
Claude Opus 5.5
claude-opus-5-5
100/ 100Excellent
Thinking signature: passed
Totals of the 3 runs: 100 · 100 · 100
Claude Fable 5.1
claude-fable-5-1
100/ 100Excellent
Thinking signature: passed
Totals of the 3 runs: 98 · 100 · 100
Claude Sonnet 5.5
claude-sonnet-5-5
100/ 100Excellent
Thinking signature: passed
Totals of the 3 runs: 100 · 100 · 100
Claude Haiku 5.5
claude-haiku-5-5
100/ 100Excellent
Thinking signature: passed
Totals of the 3 runs: 98 · 100 · 100
Score by item
| Item | Weight | Opus 5.5 | Fable 5.1 | Sonnet 5.5 | Haiku 5.5 |
|---|---|---|---|---|---|
| Authenticity | |||||
| Identity | 5 | 100 | 100 | 100 | 100 |
| Behavioral signature | 15 | 100 | 100 | 100 | 100 |
| Thinking signature | 25 | 100 | 100 | 100 | 100 |
| Model consistency | 10 | 100 | 100 | 100 | 100 |
| Knowledge | 10 | 100 | 100 | 100 | 100 |
| Capability | |||||
| PDF reading | 8 | 100 | 100 | 100 | 100 |
| Structured output | 12 | 100 | 100 | 100 | 100 |
| Protocol | |||||
| Protocol | 5 | 100 | 100 | 100 | 100 |
| Response integrity | 5 | 100 | 100 | 100 | 100 |
| Message IDs | 5 | 100 | 100 | 100 | 100 |
| Token usage | 10 | 100 | 100 | 100 | 100 |
| Total | 100 | 100 | 100 | 100 | |
The total is the weighted average of the item scores; skipped items are left out. 85 or more counts as Excellent, 70 to 84 as Pass.