api.river.ai probe report
This report aggregates BazaarLink Probe measurements for this endpoint. They describe specific points in time, not an overall service rating. A relay can change its upstream at any time, so old measurements may not reflect current behavior.
At least one probe returned behavior inconsistent with the claimed model. One occurrence is enough to flag because substitution need not happen every time.
Verdict timeline
The latest probe results for each model, oldest on the left. A switch from “match” to “swapped” is flagged in red — this is how you catch a relay that swaps models now and then, or degrades after a while. Hover a swatch for its time.
| Claimed model | Recent probes (oldest → newest) | Swap rate | Latest |
|---|---|---|---|
| deepseek-ai/DeepSeek-V4.1-Flash-1017 | ✓✓✓✓◐✓✓✓✓✓✓◐✓✓✓✓✓✓ | 0% (0/18) | Match 10/29 10:26 |
| deepseek-ai/DeepSeek-V4.1-Flash | ✕✓✓✓✓✓✓✓ | 13% (1/8) | Match 10/28 23:27 |
| moonshotai/Kimi-K3 | ◌◌◌◌◌◌◌ | 0% (0/7) | Reference being built 10/29 03:48 |
| zai-org/GLM-5.3-Flash | ◐◐◐◐ | 0% (0/4) | Same family · model unconfirmed 10/26 09:57 |
| GLM-5.3 | ✓✓✓ | 0% (0/3) | Match 10/29 04:15 |
Per-model probe results
We intentionally do not average across models: an endpoint serving 40 models is not comparable to one serving 3.
| Claimed model | Fingerprint verdict | Observed family | Probe runs | Last probe |
|---|---|---|---|---|
| deepseek-ai/DeepSeek-V4.1-Flash-1017 | Substitution | deepseek | 10 | 2026-10-28 |
| moonshotai/Kimi-K3 | Substitution | anthropic | 5 | 2026-10-28 |
| GLM-5.3 | Match | zhipu | 1 | 2026-10-27 |
| zai-org/GLM-5.3-Flash | Substitution | zhipu | 1 | 2026-10-26 |
How these verdicts are produced
Automated probes compare behavioral fingerprints, token accounting, hidden system-prompt behavior, and streaming format. Endpoint and model self-claims are not trusted; verdicts use behavioral evidence and abstain when evidence is insufficient. Relay Probe
Other endpoints claiming these models
Sorted by the number of claimed models shared with this endpoint. The same model can receive different verdicts at different endpoints.
BazaarLink also provides AI API services and is therefore a market participant. We publish the logic and reference data so you can inspect the method. If you operate this endpoint and believe a measurement is wrong, please contact us We will mark it clarified or remove it after verification.