SOURCE CATALOG HISTORY · 18–28 OCT 2026
Ten days of AI catalog change.
Two anonymous captures of the same public OpenRouter model-catalog endpoint, using the same output-modality filter, show how quickly the listed landscape changes. Source API and field reference ↗
686 → 692 LISTED MODELS · 2 HASH-CHECKED SNAPSHOTS
What changed in the listing?
| Observed field | 18 Oct | 28 Oct | Change |
|---|---|---|---|
| Listed model IDs | 686 | 692 | +6 net |
| IDs first seen in later snapshot | — | — | 24 |
| IDs absent from later snapshot | — | — | 18 |
| Existing entries with metadata changes | — | — | 81 |
| Updated entries with reported price fields changed | — | — | 52 |
| Updated entries with capability metadata changed | — | — | 15 |
The price and capability counts are overlapping subsets of the 81 updated entries. A catalog addition or removal means only that the ID appeared or did not appear in these two source responses; it does not prove launch, shutdown or live route availability.
What the newly observed entries describe
Among 24 IDs first observed in the later snapshot, source metadata lists 13 authors. Twenty-two entries list text input, 15 image input, five embeddings output and 10 the tools parameter. Input and output categories overlap. The profile summarizes declared fields; it does not establish working capability, availability or quality.
Explore all 24 newly observed entries
Context size, input/output modalities and supported tool parameter come from the source listing. Categories overlap; variants and aliases can each have their own ID. Every row belongs to this exact two-snapshot window.
Open this section to load the hash-checked profile.
Inspect the complete 24-entry profile and its source hashes ↗
A few visible examples
Newly listed IDs included inception/mercury-3, liquid/lfm-2.5-2.6b and minimax/minimax-m3.1-flash. Reported price metadata changed for entries including deepseek/deepseek-chat, deepseek/deepseek-v4-flash-latest and moonshotai/kimi-latest.
These are examples selected from the captured diff, not recommendations. The snapshots do not show which resource performs best, whether a route is reachable now, or what a real request would cost.
Evidence and limits
Explore all 52 reported price-field changes
The source reports USD per token. Values below convert that metadata to USD per million input or output tokens. The table shows prompt and completion fields; cache, image and other pricing fields remain in the full diff. Some rows are multimodal or audio resources, so token units and task semantics may differ: do not read these as like-for-like cost comparisons. A zero is only the reported field value. These are reported listing prices, not verified charges. OpenRouter Models API field reference ↗
Open this section to load the hash-verified table.
Before snapshot: 18 Oct 2026, 16:54:10 UTC · 686 IDs · model-set SHA-256 39589dbcff23…54c56104c. After snapshot: 28 Oct 2026, 13:33:48 UTC · 692 IDs · model-set SHA-256 b4553a3a2a7d…95f9fa54b. The comparison checks the source and filter match and recomputes field-level differences.
This is a change in public catalog metadata only. It does not verify actual route pricing, runtime availability, quality, latency, adoption or market share. No model inference was performed for this report.
Inspect the complete machine-readable diff ↗
Diff file SHA-256: 8cdfec0c98405394a1b5c8c611c520ca3bd4757a4a56352d62241e43be2a7c5b · Follow the live verified catalog feed ↗ · Browse measured studies ↗