Jev compared with the models you would otherwise use
Jev bills 0.042 dollars per million input tokens and nothing for output. An LLM bills both. These pages work the gap out per model, and each one says plainly where the LLM is still the better choice.
Luna is the cheapest frontier-family LLM most teams reach for when classifying. Here is what the same workload costs on each, with output tokens counted.
GPT-5.6 Luna: $0.2/M in · $1.2/M out | Jev 1.13: $0.042/M in · free out
Haiku is the usual baseline for high-volume classification and routing. LiteLLM measured Jev classifying 5.43x faster at 96% lower cost on their router corpus.
Claude Haiku 4.5: $1/M in · $5/M out | Jev 1.13: $0.042/M in · free out
Sonnet is the mid-tier default in a lot of pipelines. On pure classification volume the gap is mostly output tokens, which Jev does not bill at all.
Claude Sonnet 5: $2/M in · $10/M out | Jev 1.13: $0.042/M in · free out
The frontier end of the comparison. This is where TypeSafe's headline savings ratios come from, so treat them as an upper bound rather than a forecast.
GPT-6 Astra: $10/M in · $50/M out | Jev 1.13: $0.042/M in · free out