Models for roleplay
This is a curated list, not a filter. The other facets apply a rule to the whole catalog — a price ceiling, a context floor — and this one cannot: “good at roleplay” is a judgement, so the membership and the order are ours, and we would rather label that than dress it up as a computation.
The roleplay cut
1 model
| Input | Providers | |||||
|---|---|---|---|---|---|---|
| 128K | 8K | $0.756 | $0.756 | text |
What this page does not claim
No latency, throughput or availability figures appear above, because we have not run long enough to have measured any for these models. When we have, they will appear here as our own numbers with the date they were taken — and until then their absence is the honest state, not an oversight. Context windows and rates are live catalog values.
Some of these have nowhere to fail over to. Llama 3.3 70B Instruct is served by a single registered upstream right now. Multi-provider failover is the reliability argument for everything else we run, and it does not apply here: if that upstream has an outage, the model is simply unavailable for its duration. It stays on this list because it earns its place on output quality — but you should know what you are choosing. How failover works.
Not currently in the catalog. nousresearch/hermes-4-70b, sao10k/l3.3-euryale-70b, mistral/mistral-small-3.2 are on our candidate list but do not resolve in the live catalog, so they are not listed above. We would rather show you a short list that works than a long one that half-404s.
Client setup for SillyTavern, RisuAI and Agnai — and what we store, which is not your prompts — is on /use/roleplay. The arithmetic for what one message costs at these rates is in the cost guide.