diff --git a/providers/routing-run/models/route/glm-5.1-6bit.toml b/providers/routing-run/models/route/glm-5.1-6bit.toml index 9a0e7f914..59c7d98fc 100644 --- a/providers/routing-run/models/route/glm-5.1-6bit.toml +++ b/providers/routing-run/models/route/glm-5.1-6bit.toml @@ -13,4 +13,4 @@ output = 3.00 [limit] context = 202_752 -output = 202_752 +output = 65_536 diff --git a/providers/routing-run/models/route/glm-5.1.toml b/providers/routing-run/models/route/glm-5.1.toml index d972be3d6..dd49c9817 100644 --- a/providers/routing-run/models/route/glm-5.1.toml +++ b/providers/routing-run/models/route/glm-5.1.toml @@ -13,4 +13,4 @@ output = 3.00 [limit] context = 202_752 -output = 202_752 +output = 65_536 diff --git a/providers/routing-run/models/route/mimo-v2.5.toml b/providers/routing-run/models/route/mimo-v2.5.toml new file mode 100644 index 000000000..cde09d8bd --- /dev/null +++ b/providers/routing-run/models/route/mimo-v2.5.toml @@ -0,0 +1,21 @@ +name = "MiMo V2.5" +attachment = true +reasoning = true + +[extends] +from = "xiaomi/mimo-v2.5-pro" + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.45 +output = 1.35 + +[limit] +context = 1_000_000 +output = 262_144 + +[modalities] +input = ["text", "image", "video"] +output = ["text"] diff --git a/providers/routing-run/models/route/qwen3.6-27b-202k.toml b/providers/routing-run/models/route/qwen3.6-27b-202k.toml new file mode 100644 index 000000000..93450c6bd --- /dev/null +++ b/providers/routing-run/models/route/qwen3.6-27b-202k.toml @@ -0,0 +1,18 @@ +name = "Qwen3.6 27B 202K" +attachment = false +reasoning = false + +[extends] +from = "alibaba/qwen3.6-27b" + +[cost] +input = 1.10 +output = 3.30 + +[limit] +context = 202_000 +output = 32_768 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/providers/routing-run/models/route/qwen3.6-27b.toml b/providers/routing-run/models/route/qwen3.6-27b.toml index d928c20e3..c8d9bec25 100644 --- a/providers/routing-run/models/route/qwen3.6-27b.toml +++ b/providers/routing-run/models/route/qwen3.6-27b.toml @@ -1,4 +1,5 @@ name = "Qwen3.6 27B" +attachment = false reasoning = false [extends] @@ -9,9 +10,9 @@ input = 1.10 output = 3.30 [limit] -context = 262_144 -output = 262_144 +context = 202_000 +output = 32_768 [modalities] -input = ["text", "image", "video"] +input = ["text"] output = ["text"]