English examples; experimental phrase-level switching demonstrated for French, Spanish, German, Japanese, Chinese and KoreanSource for Languages (opens in a new tab)
These are documented examples, not a complete supported-language inventory. S2.1 Pro’s 83-language count is not inherited.
Text chunks 100–300 characters; min chunk 0–100. max_new_tokens defaults to 1,024 audio tokens per chunk. Temperature/top_p 0–1; latency normal/balanced/lowSource for Generation controls (opens in a new tab)
Chunk controls are not total input or output limits. Internal batching is not a public batch API.
Thresholds measure total prepaid funds, not spend. All keys and ASR share the pool; 429 responses have no Retry-After header. Free-tier fair-use conditions also apply.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Fish Audio terms permit Usage Data and Content to improve its models, including third-party components. Public terms do not establish universal zero retention. Free S2.1 has no DPA or TTFA guarantees; confirm production contractual terms separately. Voice inputs require appropriate rights.
Thresholds measure total prepaid funds, not spend. All keys and ASR share the pool; 429 responses have no Retry-After header. Free-tier fair-use conditions also apply.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Fish Audio terms permit Usage Data and Content to improve its models, including third-party components. Public terms do not establish universal zero retention. Free S2.1 has no DPA or TTFA guarantees; confirm production contractual terms separately. Voice inputs require appropriate rights.
Thresholds measure total prepaid funds, not spend. All keys and ASR share the pool; 429 responses have no Retry-After header. Free-tier fair-use conditions also apply.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Fish Audio terms permit Usage Data and Content to improve its models, including third-party components. Public terms do not establish universal zero retention. Free S2.1 has no DPA or TTFA guarantees; confirm production contractual terms separately. Voice inputs require appropriate rights.
No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
The official API documents this named managed model under Fish Audio service terms that reserve service IP and restrict reverse engineering. The reviewed official weight inventory does not publish this release; the separately downloadable S2 Pro and S1-mini are excluded. Exact internal checkpoint lineage remains unverified.
