Skip to content

Voice-to-text dictation app with local (Nvidia Parakeet/Whisper) and cloud models (BYOK). Privacy-first and available cross-platform.

feat: expanded qwen and groq listed models

Gabriel Stein70843877be65
diff --git a/src/models/ModelRegistry.ts b/src/models/ModelRegistry.ts
index 8b10992ea..23db4b698 100644
--- a/src/models/ModelRegistry.ts
+++ b/src/models/ModelRegistry.ts
@@ -193,13 +193,16 @@ export function getModelProvider(modelId: string): string {
if (modelId.includes("claude")) return "anthropic";
// Check for gemini but exclude gemma (which could be Groq or local)
if (modelId.includes("gemini") && !modelId.includes("gemma")) return "gemini";
- if (modelId.includes("gpt-4") || modelId.includes("gpt-5")) return "openai";
+ // OpenAI cloud models (gpt-4.x, gpt-5.x but NOT gpt-oss which could be Groq)
+ if ((modelId.includes("gpt-4") || modelId.includes("gpt-5")) && !modelId.includes("gpt-oss")) return "openai";
// Groq-specific model patterns (these run on Groq cloud, not local)
- // Groq models have patterns like "qwen/", "llama3-", "mixtral-", "gemma2-"
- if (modelId.includes("qwen/") || modelId.includes("llama3-") ||
+ // Groq models have patterns like "qwen/", "openai/gpt-oss-", "llama-3.1-8b-instant", "llama-3.3-", "mixtral-", "gemma2-"
+ if (modelId.includes("qwen/") || modelId.includes("openai/") ||
+ modelId.includes("llama-3.1-8b-instant") || modelId.includes("llama-3.3-") ||
modelId.includes("mixtral-") || modelId.includes("gemma2-")) return "groq";
- // Local models have patterns like "qwen2.5-", "llama-3.2-", "mistral-"
- if (modelId.includes("qwen") || modelId.includes("llama") || modelId.includes("mistral")) return "local";
+ // Local models have patterns like "qwen2.5-", "qwen3-", "llama-3.2-", "mistral-", "gpt-oss-20b-mxfp4"
+ if (modelId.includes("qwen") || modelId.includes("llama") ||
+ modelId.includes("mistral") || modelId.includes("gpt-oss-20b-mxfp4")) return "local";
}
return model?.provider || "openai";
diff --git a/src/models/modelRegistryData.json b/src/models/modelRegistryData.json
index f713ed80b..730deb878 100644
--- a/src/models/modelRegistryData.json
+++ b/src/models/modelRegistryData.json
@@ -90,22 +90,27 @@
{
"id": "qwen/qwen3-32b",
"name": "Qwen3 32B",
- "description": "Powerful reasoning model"
+ "description": "Powerful reasoning with thinking mode, 131K context"
},
{
- "id": "llama-3.3-70b-versatile",
- "name": "LLaMA 3.3 70B",
- "description": "Meta's latest versatile model"
+ "id": "openai/gpt-oss-120b",
+ "name": "GPT-OSS 120B",
+ "description": "OpenAI's open-source flagship, 500 T/sec"
+ },
+ {
+ "id": "openai/gpt-oss-20b",
+ "name": "GPT-OSS 20B",
+ "description": "Fast open-source model, 1000 T/sec"
},
{
- "id": "llama3-70b-8192",
- "name": "LLaMA 3 70B",
- "description": "8K context, balanced performance"
+ "id": "llama-3.3-70b-versatile",
+ "name": "LLaMA 3.3 70B",
+ "description": "Meta's versatile model, 280 T/sec"
},
{
- "id": "llama3-8b-8192",
- "name": "LLaMA 3 8B",
- "description": "Fast and efficient"
+ "id": "llama-3.1-8b-instant",
+ "name": "LLaMA 3.1 8B",
+ "description": "Ultra-fast 560 T/sec, 131K context"
},
{
"id": "mixtral-8x7b-32768",
@@ -121,9 +126,84 @@
}
],
"localProviders": [
+ {
+ "id": "qwen3",
+ "name": "Qwen3",
+ "baseUrl": "https://huggingface.co",
+ "promptTemplate": "<|im_start|>system\n{system}<|im_end|>\n<|im_start|>user\n{user}<|im_end|>\n<|im_start|>assistant\n",
+ "models": [
+ {
+ "id": "qwen3-8b-q4_k_m",
+ "name": "Qwen3 8B",
+ "size": "5.0GB",
+ "sizeBytes": 5368709120,
+ "description": "Latest Qwen3 with thinking mode support",
+ "fileName": "qwen3-8b-q4_k_m.gguf",
+ "quantization": "q4_k_m",
+ "contextLength": 131072,
+ "hfRepo": "Qwen/Qwen3-8B-GGUF",
+ "recommended": true
+ },
+ {
+ "id": "qwen3-8b-q5_k_m",
+ "name": "Qwen3 8B (Q5)",
+ "size": "5.9GB",
+ "sizeBytes": 6333399654,
+ "description": "Higher quality Qwen3 with thinking mode",
+ "fileName": "qwen3-8b-q5_k_m.gguf",
+ "quantization": "q5_k_m",
+ "contextLength": 131072,
+ "hfRepo": "Qwen/Qwen3-8B-GGUF"
+ },
+ {
+ "id": "qwen3-4b-q4_k_m",
+ "name": "Qwen3 4B",
+ "size": "2.5GB",
+ "sizeBytes": 2684354560,
+ "description": "Compact Qwen3 with reasoning capabilities",
+ "fileName": "qwen3-4b-q4_k_m.gguf",
+ "quantization": "q4_k_m",
+ "contextLength": 131072,
+ "hfRepo": "Qwen/Qwen3-4B-GGUF"
+ },
+ {
+ "id": "qwen3-1.7b-q8_0",
+ "name": "Qwen3 1.7B",
+ "size": "1.8GB",
+ "sizeBytes": 1932735283,
+ "description": "Small but capable Qwen3 model",
+ "fileName": "qwen3-1.7b-q8_0.gguf",
+ "quantization": "q8_0",
+ "contextLength": 131072,
+ "hfRepo": "Qwen/Qwen3-1.7B-GGUF"
+ },
+ {
+ "id": "qwen3-0.6b-q8_0",
+ "name": "Qwen3 0.6B",
+ "size": "0.6GB",
+ "sizeBytes": 644245094,
+ "description": "Tiny Qwen3 for edge devices",
+ "fileName": "qwen3-0.6b-q8_0.gguf",
+ "quantization": "q8_0",
+ "contextLength": 131072,
+ "hfRepo": "Qwen/Qwen3-0.6B-GGUF"
+ },
+ {
+ "id": "qwen3-32b-q4_k_m",
+ "name": "Qwen3 32B",
+ "size": "19.8GB",
+ "sizeBytes": 21260251955,
+ "description": "Most powerful local Qwen3 with deep reasoning",
+ "fileName": "qwen3-32b-q4_k_m.gguf",
+ "quantization": "q4_k_m",
+ "contextLength": 131072,
+ "hfRepo": "Qwen/Qwen3-32B-GGUF"
+ }
+ ]
+ },
{
"id": "qwen",
- "name": "Qwen",
+ "name": "Qwen 2.5",
"baseUrl": "https://huggingface.co",
"promptTemplate": "<|im_start|>system\n{system}<|im_end|>\n<|im_start|>user\n{user}<|im_end|>\n<|im_start|>assistant\n",
"models": [
@@ -169,8 +249,7 @@
"fileName": "qwen2.5-7b-instruct-q4_k_m.gguf",
"quantization": "q4_k_m",
"contextLength": 128000,
- "hfRepo": "Qwen/Qwen2.5-7B-Instruct-GGUF",
- "recommended": true
+ "hfRepo": "Qwen/Qwen2.5-7B-Instruct-GGUF"
},
{
"id": "qwen2.5-7b-instruct-q5_k_m",