diff --git a/frontend/src/lib/components/Editor.svelte b/frontend/src/lib/components/Editor.svelte
index 552315a28d..d24ed3e724 100644
--- a/frontend/src/lib/components/Editor.svelte
+++ b/frontend/src/lib/components/Editor.svelte
@@ -463,7 +463,7 @@
 				async provideInlineCompletionItems(model, position, context, token) {
 					abortController?.abort()
 					const textUntilPosition = model.getText(
-						new vscode.Range(1, 1, position.line, position.character)
+						new vscode.Range(0, 0, position.line, position.character)
 					)
 					let items: vscode.InlineCompletionItem[] = []
 
diff --git a/frontend/src/lib/components/copilot/completion.ts b/frontend/src/lib/components/copilot/completion.ts
index 8eccd591fb..e8a669d5e1 100644
--- a/frontend/src/lib/components/copilot/completion.ts
+++ b/frontend/src/lib/components/copilot/completion.ts
@@ -2,24 +2,15 @@ import type { ChatCompletionMessageParam } from 'openai/resources/chat'
 import { getNonStreamingCompletion } from './lib'
 import { codeCompletionLoading } from '$lib/stores'
 
-const systemPrompt =
-	'You are a code completion assistant, return the code that should go inside the <completion></completion> tags. If you add a line break, take into account the indentation of the code. You can also not return anything if you think the code is already complete.'
-const prompt = `complete the following code:
-\`\`\`{language}
-{before}<completion></completion>{after}
-\`\`\``
+const systemPrompt = `You are a code completion assistant, return the code that should go instead of the <completion_tokens>.
 
-const exampleUser = prompt
-	.replace('{language}', 'python')
-	.replace(
-		'{before}',
-		`def main(number):
-    # print the number and divide it by 4
-    `
-	)
-	.replace('{after}', `\n\n    return number`)
-const exampleAssistant = `<completion>print(number)
-		number = number / 4</completion>`
+Only return the completion tokens. Wrap the completion tokens in a code block (\`\`\`lang\n[completion_tokens]\n\`\`\`).
+
+Return \`None\` if you think the code is already complete.`
+const prompt = `\`\`\`{language}
+{before}<completion_tokens>{after}
+\`\`\`
+`
 
 export async function editorCodeCompletion(
 	before: string,
@@ -33,14 +24,6 @@ export async function editorCodeCompletion(
 			role: 'system',
 			content: systemPrompt
 		},
-		{
-			role: 'user',
-			content: exampleUser
-		},
-		{
-			role: 'assistant',
-			content: exampleAssistant
-		},
 		{
 			role: 'user',
 			content: prompt
@@ -51,8 +34,17 @@ export async function editorCodeCompletion(
 	]
 
 	try {
-		const result = await getNonStreamingCompletion(messages, abortController, 'gpt-3.5-turbo')
-		const completion = result.match(/<completion>(.*)<\/completion>/s)?.[1] || ''
+		const result = await getNonStreamingCompletion(messages, abortController, 'gpt-3.5-turbo-1106')
+
+		const match = result.match(/```[a-zA-Z]+\n([\s\S]*?)\n```/)
+
+		let completion = match?.[1] || ''
+		if (completion) {
+			const previousIndent = before.match(/(\n|^)([ \t]*)$/)?.[2]
+			if (previousIndent) {
+				completion = completion.replace(/\n/g, `\n${previousIndent}`)
+			}
+		}
 
 		return completion
 	} catch (err) {
diff --git a/frontend/src/lib/components/copilot/lib.ts b/frontend/src/lib/components/copilot/lib.ts
index 85c7b365e6..78a21c9664 100644
--- a/frontend/src/lib/components/copilot/lib.ts
+++ b/frontend/src/lib/components/copilot/lib.ts
@@ -16,8 +16,8 @@ export const SUPPORTED_LANGUAGES = new Set(Object.keys(GEN_CONFIG.prompts))
 
 const openaiConfig: ChatCompletionCreateParamsStreaming = {
 	temperature: 0,
-	max_tokens: 2048,
-	model: 'gpt-4-1106-preview',
+	max_tokens: 4096,
+	model: 'gpt-4-0125-preview',
 	seed: 42,
 	stream: true,
 	messages: []