AllocateThe cheapest language model in every context tier
Paying for context you never send is the quiet waste in model selection. The cheapest model in each tier: Qwen 2 Instruct (1.5B) (32K) at $0.02, Trinity Mini (128K) at $0.045, Qwen3 Next 80B A3b Instruct (256K) at $0.15, MiniMax M3 (512K) at $0.30, Llama 4 Scout Instruct (17Bx16E) (1M) at $0.18 per million input tokens at list.
Provider list prices from the Allocate catalog, checked 2026-07-08. Billed price on Allocate is list plus the 7% transaction fee.
Match the window to the workload
Requests have a natural size: a support turn is thousands of tokens, a contract is tens of thousands, a corpus is hundreds of thousands. The right model is the cheapest one whose window fits your real request size, not the biggest window on the menu.
Each tier lists its two cheapest models by input list price. Ranked this way, the price of context itself becomes visible: the jump from a 128K-class model to a 1M-class one is often smaller than the jump between two labs in the same tier.
Common questions
How do I pick a context tier?
Measure your real request sizes (the token counter helps) and pick the tier that fits the 95th percentile, not the maximum imaginable document. Route the rare oversized request to a bigger-window model instead of paying its price on every call.
What is the cheapest 1M-context model?
Llama 4 Scout Instruct (17Bx16E) at $0.18 per million input tokens at list.
Related
Every model here sits behind one key on Allocate: route by name, meter per route, and swap the model in one click.