fix(inference): point MiniMax at its OpenAI-compatible /v1 endpoint (#3401) (#3402)

This commit is contained in:
oxoxDev
2026-06-08 18:26:49 +05:30
committed by GitHub
parent 04629ea16e
commit 4c734afe6d
5 changed files with 29 additions and 12 deletions
+12 -4
View File
@@ -132,8 +132,16 @@ pub const BUILTIN_CLOUD_PROVIDERS: &[BuiltinCloudProvider] = &[
BuiltinCloudProvider {
slug: "minimax",
label: "MiniMax",
endpoint: "https://api.minimax.io/anthropic",
auth_style: AuthStyle::Anthropic,
// MiniMax exposes a full OpenAI-compatible surface at `/v1`
// (`/v1/chat/completions`, `/v1/models`). The previous `/anthropic`
// base + Anthropic auth pointed at MiniMax's Messages-protocol API,
// which OpenHuman does not speak — it only builds OpenAI-style
// `/chat/completions` and `/models` — so both chat and model-listing
// 404'd (`/anthropic/chat/completions`, `/anthropic/models`). The
// 404 on model-listing was Sentry TAURI-RUST-8X3. Use the `/v1`
// OpenAI surface with Bearer auth so both paths resolve.
endpoint: "https://api.minimax.io/v1",
auth_style: AuthStyle::Bearer,
},
BuiltinCloudProvider {
slug: "stepfun",
@@ -507,8 +515,8 @@ mod tests {
(
"minimax",
"MiniMax",
"https://api.minimax.io/anthropic",
AuthStyle::Anthropic,
"https://api.minimax.io/v1",
AuthStyle::Bearer,
),
(
"sumopod",