- Free allowance
- 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).
- Verdict
- caveatusable, with a named limit or an unprobed read path
- Cost
- free tier
- Identity
- card?placeholder, not evidence; card unknown, phone unknown
- Binding limit
- per-object-capper-object-cap: NOT published, and NOT 2 GiB
- Evidence
- documented8 sources; a first-party page was read and the sentence quoted
- Freshness
- stale on 2026-11-21checked 2026-10-21 (inherited); stale on 2026-11-21
| 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 |
|---|
| Limits | - per-object-cap (binding) per-object-cap: NOT published, and NOT 2 GiB source
per-object-cap: NOT published, and NOT 2 GiB like GitHub — verified live at 489,088,538,784 bytes in one blob. 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. |
|---|
| Identity | card (inferred, not read) not published |
|---|
| 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. |
|---|
| 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 |
|---|
| 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. |
|---|
| Evidence note | documented The page states the number in the record; the claim is documented, not measured. |
|---|
| Parsed numbers | - has-free-tier 1
- per-object-cap 2 GiB
|
|---|