Add Perplexity reasoning pro. Add Groq Llama 70b R1 distill

Author: danielcampagnolitg

src/functions/web/perplexity.ts

@@ -41,16 +41,13 @@ export class Perplexity {
 
 			// Valid model ids are at https://docs.perplexity.ai/guides/model-cards
 			response = await perplexity.chat.completions.create({
-				model: 'sonar-pro',
-				max_tokens: 4096,
+				model: 'sonar-reasoning-pro',
 				messages: [{ role: 'user', content: researchQuery }],
 				stream: false,
 			});
 			const content = response.choices[0].message?.content;
 
-			if (!content) {
-				throw new Error('Perplexity API returned empty content');
-			}
+			if (!content) throw new Error('Perplexity API returned empty content');
 
 			// Cost calculation based on Perplexity API pricing (as of the last update)
 			// Source: https://docs.perplexity.ai/docs/pricing
@@ -59,9 +56,9 @@ export class Perplexity {
 				const completionTokens = response.usage.completion_tokens;
 				const totalTokens = response.usage.total_tokens;
 
-				const costPerPromptToken = 0.000001; // $1 per million tokens
-				const costPerCompletionToken = 0.000001; // $1 per million tokens
-				const onlineCost = 0.005; // $5 per 1000 requests
+				const costPerPromptToken = 0.000002; // $1 per million tokens
+				const costPerCompletionToken = 0.000008; // $1 per million tokens
+				const onlineCost = 0.015; // $5 per 1000 requests. 3 requests
 
 				const cost = Number((promptTokens * costPerPromptToken + completionTokens * costPerCompletionToken + onlineCost).toFixed(6));
 				addCost(cost);
@@ -72,7 +69,7 @@ export class Perplexity {
 			if (saveToMemory) {
 				const summary = await llms().easy.generateText(
 					`<query>${researchQuery}</query>\nGenerate a summarised version of the research key in one short sentence at most, with only alphanumeric with underscores for spaces. Answer concisely with only the summary.`,
-					{ id: 'summarisePerplexityQuery' },
+					{ id: 'Summarise Perplexity search' },
 				);
 				const key = `Perplexity-${summary}`;
 				agentContext().memory[key] = content;

src/llm/services/groq.ts

@@ -9,6 +9,7 @@ export const GROQ_SERVICE = 'groq';
 export function groqLLMRegistry(): Record<string, () => LLM> {
 	return {
 		'groq:llama-3.3-70b-versatile': groqLlama3_3_70B,
+		'groq:deepseek-r1-distill-llama-70b': groqLlama3_3_70B_R1_Distill,
 	};
 }
 
@@ -20,6 +21,10 @@ export function groqLlama3_3_70B(): LLM {
 	return new GroqLLM('Llama3.3 70b (Groq)', 'llama-3.3-70b-versatile', 131_072, perMilTokens(0.59), perMilTokens(0.79));
 }
 
+export function groqLlama3_3_70B_R1_Distill(): LLM {
+	return new GroqLLM('Llama3.3 70b R1 Distill (Groq)', 'deepseek-r1-distill-llama-70b', 1280_000, perMilTokens(0.59), perMilTokens(0.79));
+}
+
 /**
  * https://wow.groq.com/
  */

src/llm/services/perplexity-llm.ts

@@ -15,15 +15,17 @@ export const PERPLEXITY_SERVICE = 'perplexity';
 
 /*
 https://docs.perplexity.ai/guides/pricing
-Model	                            Price per 1000 requests	   Price per 1M tokens
-llama-3.1-sonar-large-128k-online	$5	                       $1
-llama-3.1-sonar-huge-128k-online	$5	                       $5
+Model	Input Tokens (Per Million Tokens)	Output Tokens (Per Million Tokens)	Price per 1000 searches
+sonar-reasoning-pro	$2	$8	$5
+sonar-reasoning	$1	$5	$5
+sonar-pro	$3	$15	$5
+sonar	$1	$1	$5
 */
 
 export function perplexityLLMRegistry(): Record<string, () => LLM> {
 	return {
 		[`${PERPLEXITY_SERVICE}:sonar`]: perplexityLLM,
-		[`${PERPLEXITY_SERVICE}:sonar-pro`]: perplexityProLLM,
+		[`${PERPLEXITY_SERVICE}:sonar-reasoning-pro`]: perplexityReasoningProLLM,
 	};
 }
 
@@ -38,24 +40,13 @@ export function perplexityLLM(): LLM {
 	);
 }
 
-export function perplexityProLLM(): LLM {
+export function perplexityReasoningProLLM(): LLM {
 	return new PerplexityLLM(
-		'Perplexity Pro',
-		'sonar-pro',
-		200_000, // maxTokens
-		0.000005, // costPerPromptToken ($5 per million tokens)
-		0.000005, // costPerCompletionToken
-		0.015, // 3 searches ($5 per 1000 requests)
-	);
-}
-
-export function perplexityReasoningLLM(): LLM {
-	return new PerplexityLLM(
-		'Perplexity Reasoning',
-		'sonar-reasoning',
+		'Perplexity Reasoning Pro',
+		'sonar-reasoning-pro',
 		127_000, // maxTokens
-		0.000001, // costPerPromptToken ($1 per million tokens)
-		0.000005, // costPerCompletionToken
+		0.000002, // costPerPromptToken ($1 per million tokens)
+		0.000008, // costPerCompletionToken
 		0.005, // 1 search ($5 per 1000 requests)
 	);
 }