Ollama registry
ollama.ai · Storage · 1 record · JSON
Ollama registry (registry.ollama.ai) — the strongest finding in this slot Storage
caveatcardcardDper-object-cap
| Free tier | No published byte quota, no published per-file cap, and no paid storage tier exists to upsell to — the only paid Ollama product is inference compute ('Pricing': Free $0 'Use any cloud model with pay-as-you-go usage credits', Pro $20/mo; nothing on that page meters model storage). A single repo holds 455.5 GiB in one layer. unit GiB; metered on stored-bytes |
|---|---|
| Price | none stated |
| Per-object cap | per-object-cap: NOT published, and NOT 2 GiB (2147483648 bytes) |
| Retention | client-side chunking is 100 MB-1 GB parts (server/upload.go). Retention: none — objects persist until deleted; no TTL anywhere in the docs |
| Egress | Unmetered anonymous pull. No egress line exists anywhere in Ollama's pricing — there is no per-GB transfer meter to quote because pull is free and unauthenticated. Blob GET is 307 to an R2 bucket (dd20bb891979d25aebc8bec07b2b3bbc.r2.cloudflarestorage.com, X-Amz-Expires=86400), fronted by Cloudflare. |
| Identity | card derived from the record's prose, not from an identity field, and no signup was executed: y paid Ollama product is inference compute ('Pricing': Free $0 'Use any cloud model with pay-as-you-go usage credits', Pro $20/mo; nothing on that page meters model storage). Empiric |
| Limits |
per-object-cap: NOT published, and NOT 2 GiB like GitHub — verified live at 489,088,538,784 bytes in one blob. Practical upload ceiling is the server's 413: a user report (issue #13740, opened 2026-01-16) shows a 109 GB push getting '413 Request Entity Too Large' after 1 GB, so ~100 GB is a plausible hard wall but is unverified by me (I have no push credentials). client-side chunking is 100 MB-1 GB parts (server/upload.go). Retention: none — objects persist until deleted; no TTL anywhere in the docs. request-rate: not published; 20 consecutive ranged GETs all returned 206 with no throttling. |
| Binding | per-object-cap the per-object cap sets the shard size and therefore the object count; it bounds the total more tightly than any headline quota |
| Interface | oci oci |
| Tooling | Native OCI Distribution v2. ollama push / ollama pull, oras/skopeo/crane all work. Push requires a free ollama.com account (WorkOS authkit signup, no payment); pull is fully anonymous. ollama create accepts safetensors dirs, GGUF, split GGUF globs, and any layer content — a pushed model is just OCI layers, so arbitrary bytes ride as application/vnd.ollama.image.model. |
| Composes with | unlimited free namespaces x unlimited repos x unlimited tags; the binding per-object ceiling is whatever the server accepts, and 455 GiB single-blob objects already exist in production. 30 TiB = 66 objects the size of huihui_ai/kimi-k2-abliterated. Multiply by free-account count. This is structurally identical to GHCR but with a ~45x larger observed per-object ceiling and no documented 10 GB layer cap. |
| Reliability | Pushes to large blobs are flaky: issue #6466 reports 502 Bad Gateway at 40-50% of an 8.5 GB upload, and #13740 reports 413 on 100 GB — resume/retry is required, and ollama push is not resumable by default. Reads were 100% stable in every probe. |
| Notes | No published byte quota, no published per-file cap, and no paid storage tier exists to upsell to — the only paid Ollama product is inference compute ('Pricing': Free $0 'Use any cloud model with pay-as-you-go usage credits', Pro $20/mo; nothing on that page meters model storage). A single repo holds 455.5 GiB in one layer. |
| Evidence rung | documented re-read from fetch-broker on 2026-10-22 (HTTP 404, 19 bytes, sha256 b16e15764b8bc06c). The page states the number in the record; the claim is documented, not measured. |
| Checked | 2026-10-21 (inherited); stale on 2026-11-20 |
| Parsed numbers |
|
Evidence (8)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | registry.ollama.ai/v2/ | the registry endpoint answers. Re-read https://registry.ollama.ai/v2/: HTTP 404, 19 bytes, sha256 b16e15764b8bc06c. Quoted: 404 page not found |
| 2026-10-21 | withheld | probe: GET registry.ollama.ai/v2/library/llama3.3/manifests/latest -> 200, 39.6 GiB total, largest layer 40,550.6 MiB |
| 2026-10-21 | withheld | probe: GET registry.ollama.ai/v2/library/deepseek-v3/manifests/latest -> largest layer 404,430,186,016 bytes (376.7 GiB) |
| 2026-10-21 | withheld | probe: GET registry.ollama.ai/v2/huihui_ai/kimi-k2-abliterated/blobs/sha256:33e8a1e4... -r 0-1023 -> 307 to R2 -> 206, Content-Range: bytes 0-1023/489088538784 (455.5 GiB, anonymous) |
| 2026-10-21 | withheld | probe: summed all 101 huihui_ai repos -> 89 resolvable, 3,270 GiB total, largest single 455 GiB (ONE free namespace) |
| 2026-10-21 | withheld | probe: POST registry.ollama.ai/v2/test/x/blobs/uploads/ anonymous -> 401 www-authenticate Bearer realm=registry.ollama.ai/v2/token (push is authenticated, pull is not) |
| 2026-10-21 | github.com/ollama/ollama/issues/13740 | '109 GB ... 413 Request Entity Too Large' (upload ceiling, medium confidence) |
| 2026-10-21 | registry.ollama.com/pricing | Free $0 / Pro $20/mo, both inference-only; no storage meter |
More fields (1)
{
"verified_in_pass": [
{
"url": "https://registry.ollama.ai/v2/",
"date": "2026-10-22",
"via": "fetch-broker",
"http_status": 404,
"bytes": 19,
"sha256": "b16e15764b8bc06c5c3f9f19bc8b99fa48e7894aa5a6ccdad65da49bbf564793",
"pattern_found": true,
"what_changed": "no change: the page still states the claim",
"capture": "captures\\ollama-registry-registry-ollama-ai-v2-.txt"
}
]
}Unknown or withheld values (10)
$.evidence[1].urlprivate-url$.evidence[2].urlprivate-url$.evidence[3].urlprivate-url$.evidence[4].urlprivate-url$.evidence[5].urlprivate-url$.free_tier.notesprocess$.free_tier.statedpartly-held$.notespartly-held$.per_object_cap.notesnot-extracted$.retention.notesnot-extracted