diff --git a/providers/inferx/logo.svg b/providers/inferx/logo.svg new file mode 100644 index 000000000..83899681a --- /dev/null +++ b/providers/inferx/logo.svg @@ -0,0 +1,5 @@ + + + + + diff --git a/providers/inferx/models/google/gemma-4-31b-it-fp8.toml b/providers/inferx/models/google/gemma-4-31b-it-fp8.toml new file mode 100644 index 000000000..11ac95341 --- /dev/null +++ b/providers/inferx/models/google/gemma-4-31b-it-fp8.toml @@ -0,0 +1,21 @@ +name = "Gemma 4 31B IT FP8" +family = "gemma" +release_date = "2026-05" +last_updated = "2026-05-05" +attachment = false +reasoning = false +temperature = true +tool_call = true +open_weights = true + +[cost] +input = 0.0 +output = 0.0 + +[limit] +context = 262144 +output = 8192 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/providers/inferx/models/qwen/qwen3-coder-next-fp8-1m.toml b/providers/inferx/models/qwen/qwen3-coder-next-fp8-1m.toml new file mode 100644 index 000000000..2eae421aa --- /dev/null +++ b/providers/inferx/models/qwen/qwen3-coder-next-fp8-1m.toml @@ -0,0 +1,21 @@ +name = "Qwen3 Coder Next FP8 1M" +family = "qwen" +release_date = "2026-05" +last_updated = "2026-05-12" +attachment = false +reasoning = true +temperature = true +tool_call = true +open_weights = true + +[cost] +input = 0.0 +output = 0.0 + +[limit] +context = 1024000 +output = 8192 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/providers/inferx/models/qwen/qwen3-coder-next-fp8.toml b/providers/inferx/models/qwen/qwen3-coder-next-fp8.toml new file mode 100644 index 000000000..acf16a254 --- /dev/null +++ b/providers/inferx/models/qwen/qwen3-coder-next-fp8.toml @@ -0,0 +1,21 @@ +name = "Qwen3 Coder Next FP8" +family = "qwen" +release_date = "2026-05" +last_updated = "2026-05-08" +attachment = false +reasoning = true +temperature = true +tool_call = true +open_weights = true + +[cost] +input = 0.0 +output = 0.0 + +[limit] +context = 256144 +output = 8192 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/providers/inferx/models/qwen/qwen3.5-122b-a10b-nvfp4.toml b/providers/inferx/models/qwen/qwen3.5-122b-a10b-nvfp4.toml new file mode 100644 index 000000000..d8108cef5 --- /dev/null +++ b/providers/inferx/models/qwen/qwen3.5-122b-a10b-nvfp4.toml @@ -0,0 +1,21 @@ +name = "Qwen3.5 122B A10B NVFP4" +family = "qwen3.5" +release_date = "2026-05" +last_updated = "2026-05-13" +attachment = false +reasoning = true +temperature = true +tool_call = true +open_weights = true + +[cost] +input = 0.0 +output = 0.0 + +[limit] +context = 256144 +output = 8192 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/providers/inferx/models/qwen/qwen3.6-27b-fp8.toml b/providers/inferx/models/qwen/qwen3.6-27b-fp8.toml new file mode 100644 index 000000000..f8e026bec --- /dev/null +++ b/providers/inferx/models/qwen/qwen3.6-27b-fp8.toml @@ -0,0 +1,21 @@ +name = "Qwen3.6 27B FP8" +family = "qwen3.6" +release_date = "2026-05" +last_updated = "2026-05-13" +attachment = false +reasoning = true +temperature = true +tool_call = true +open_weights = true + +[cost] +input = 0.0 +output = 0.0 + +[limit] +context = 262144 +output = 8192 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/providers/inferx/models/qwen/qwen3.6-35b-a3b-fp8.toml b/providers/inferx/models/qwen/qwen3.6-35b-a3b-fp8.toml new file mode 100644 index 000000000..0913ddbe4 --- /dev/null +++ b/providers/inferx/models/qwen/qwen3.6-35b-a3b-fp8.toml @@ -0,0 +1,21 @@ +name = "Qwen3.6 35B A3B FP8" +family = "qwen3.6" +release_date = "2026-05" +last_updated = "2026-05-13" +attachment = false +reasoning = true +temperature = true +tool_call = true +open_weights = true + +[cost] +input = 0.0 +output = 0.0 + +[limit] +context = 262000 +output = 8192 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/providers/inferx/provider.toml b/providers/inferx/provider.toml new file mode 100644 index 000000000..1541ddb36 --- /dev/null +++ b/providers/inferx/provider.toml @@ -0,0 +1,5 @@ +name = "InferX" +env = ["INFERX_API_KEY"] +npm = "@ai-sdk/openai-compatible" +api = "https://model.inferx.net/v1" +doc = "https://inferx.net"