Pricing

List prices from the Tinker model catalogue, read directly from the vendor's machine-readable catalogue each time this site rebuilds — not a snapshot someone typed in.

Inkling 2 Small costs $0.3 per million input tokens and $1 per million output tokens on Tinker, with cached input at $0.03. Fine-tuning costs $1.5 per million training tokens at 64K context and $3 at 256K. Inference prices are the same for both context variants.

List price

Tinker catalogue

US dollars per million tokens. Prefill is input processing, sample is generation, train is fine-tuning data. Cached prefill applies when a prefix was seen recently.
Context	Prefill	Cached prefill	Sample	Train
64K	$0.3	$0.03	$1	$1.5
256K	$0.3	$0.03	$1	$3

Cached prefill is 10% of the standard prefill price, a 90% discount.

What a workload actually costs

Computed from the list prices above at the 64K variant. These are arithmetic on published rates, not measured invoices.
Workload	Cost
One 1,000-token chat reply	$0.0012
Summarising a 25,000-token document	$0.0083
An agent session: 200K in, 20K out	$0.08
Fine-tuning on 10M training tokens	$15.00

Cost against measured output

Cost per index task: $1.18 as measured by Artificial Analysis
Output tokens, index run: 154M — the model is verbose relative to its peers
Time per index task: 18.5 minutes against a 7-minute median

The list price looks cheap against closed frontier models, but the model generates a lot of tokens and takes a long time doing it. For latency-bound products the effective cost is higher than the per-token figure suggests.

Compared with its siblings

Every Inkling variant in the Tinker catalogue, read live. Inkling 2 Small is the cheapest per output token of the family at present.
Model	Params	Active	Context	Prefill	Sample
Inkling	975B	41B	64K	$1.87	$4.68
Inkling (256K)	975B	41B	256K	$1.87	$4.68
Inkling-Small	276B	12B	64K	$0.58	$1.44
Inkling-Small (256K)	276B	12B	256K	$0.58	$1.44

What this price does not include

Tinker bills separately for storage of fine-tuned weights and for sampling with adapters; the catalogue covers per-token rates only. Self-hosting has no list price at all because the weights are not public yet — for the memory cost of running it yourself, see hardware requirements.

Prices last read Nov 6, 2026. They change without notice; the Tinker catalogue is the authority.