diff --git a/utils/changelog.summary.js b/utils/changelog.summary.js
new file mode 100644
index 00000000000000..92a521a29bddce
--- /dev/null
+++ b/utils/changelog.summary.js
@@ -0,0 +1,556 @@
+import fs from 'fs';
+import path from 'path';
+import { fileURLToPath } from 'url';
+import readline from 'readline';
+import { execSync } from 'child_process';
+
+const __filename = fileURLToPath( import.meta.url );
+const __dirname = path.dirname( __filename );
+
+async function askQuestion( query ) {
+
+	const rl = readline.createInterface( {
+		input: process.stdin,
+		output: process.stdout
+	} );
+
+	return new Promise( resolve => rl.question( query, ans => {
+
+		rl.close();
+		resolve( ans );
+
+	} ) );
+
+}
+
+function getExamplesChanges( releaseNumber ) {
+
+	const added = [];
+	const modified = [];
+	const removed = [];
+
+	try {
+
+		const diff = execSync( `git diff --name-status r${releaseNumber - 1}..HEAD -- examples/`, { cwd: path.join( __dirname, '..' ), encoding: 'utf8' } );
+		const lines = diff.split( '\n' );
+
+		for ( const line of lines ) {
+
+			if ( ! line ) continue;
+
+			if ( line.startsWith( 'R' ) ) {
+
+				const parts = line.split( /\s+/ );
+				if ( parts.length >= 3 ) {
+
+					const remName = parts[ 1 ].replace( 'examples/', '' ).replace( '.html', '' );
+					const addName = parts[ 2 ].replace( 'examples/', '' ).replace( '.html', '' );
+
+					removed.push( `[${remName}](https://threejs.org/examples/#${remName})` );
+					added.push( `[${addName}](https://threejs.org/examples/#${addName})` );
+
+				}
+
+				continue;
+
+			}
+
+			const parts = line.split( /\s+/ );
+			if ( parts.length < 2 ) continue;
+
+			const status = parts[ 0 ];
+			const file = parts[ 1 ];
+
+			if ( file.endsWith( '.html' ) ) {
+
+				const name = file.replace( 'examples/', '' ).replace( '.html', '' );
+				const link = `[${name}](https://threejs.org/examples/#${name})`;
+
+				if ( status.startsWith( 'A' ) ) {
+
+					added.push( link );
+
+				} else if ( status.startsWith( 'D' ) ) {
+
+					removed.push( link );
+
+				} else if ( status.startsWith( 'M' ) ) {
+
+					modified.push( link );
+
+				}
+
+			}
+
+		}
+
+	} catch ( e ) {
+
+		console.error( 'Failed to get examples diff:', e.message );
+
+	}
+
+	let output = '';
+
+	if ( added.length > 0 ) output += `**New Examples:**\n${added.join( ', ' )}\n\n`;
+	if ( modified.length > 0 ) output += `**Modified Examples:**\n${modified.join( ', ' )}\n\n`;
+	if ( removed.length > 0 ) output += `**Removed Examples:**\n${removed.join( ', ' )}\n\n`;
+
+	return output;
+
+}
+
+function getReleaseAndCacheArgs() {
+
+	let arg = process.argv[ 2 ];
+
+	if ( ! arg || arg === '-cache' ) {
+
+		// Fallback to reading REVISION from src/constants.js
+		try {
+
+			const constantsPath = path.join( __dirname, '../src/constants.js' );
+			const constantsContent = fs.readFileSync( constantsPath, 'utf8' );
+			const revisionMatch = constantsContent.match( /export\s+const\s+REVISION\s*=\s*['"](.*?)['"]/ );
+
+			if ( revisionMatch && revisionMatch[ 1 ] ) {
+
+				arg = revisionMatch[ 1 ].replace( /dev$/, '' );
+				console.log( `No release specified. Using current REVISION: ${arg}` );
+
+				// Adjust argv indexing if the first argument was actually the -cache flag
+				if ( process.argv[ 2 ] === '-cache' ) {
+
+					process.argv.splice( 2, 0, arg ); // Shift everything over by acting as if arg was passed
+
+				}
+
+			} else {
+
+				console.error( 'Usage: npm run changelog [milestone-url-or-number] [-cache <per_page>]' );
+				console.error( 'Example: npm run changelog https://github.com/mrdoob/three.js/milestone/97 -cache 50' );
+				console.error( 'Could not find REVISION in src/constants.js.' );
+				process.exit( 1 );
+
+			}
+
+		} catch ( e ) {
+
+			console.error( 'Usage: npm run changelog [milestone-url-or-number] [-cache <per_page>]' );
+			console.error( 'Example: npm run changelog https://github.com/mrdoob/three.js/milestone/97 -cache 50' );
+			console.error( 'Error reading src/constants.js:', e.message );
+			process.exit( 1 );
+
+		}
+
+	}
+
+	let perPage = 100;
+
+	for ( let i = 2; i < process.argv.length; i ++ ) { // start from 2, just in case -cache is the first arg now
+
+		const param = process.argv[ i ];
+
+		if ( param === '-cache' && i + 1 < process.argv.length ) {
+
+			perPage = parseInt( process.argv[ i + 1 ], 10 );
+			// skip the value
+
+			if ( isNaN( perPage ) || perPage <= 0 || perPage > 100 ) {
+
+				console.error( 'Invalid -cache value. It must be a number between 1 and 100.' );
+				process.exit( 1 );
+
+			}
+
+		}
+
+	}
+
+	const match = arg.match( /(\d+)$/ );
+	if ( ! match ) {
+
+		console.error( 'Invalid format. Please provide a release number (e.g. 184 or r184).' );
+		process.exit( 1 );
+
+	}
+
+	const releaseNumber = parseInt( match[ 1 ], 10 );
+	const milestoneNumber = releaseNumber - 87;
+
+	return { releaseNumber, milestoneNumber, perPage };
+
+}
+
+async function selectGeminiModel( geminiToken ) {
+
+	console.error( '\nFetching available Gemini models...' );
+
+	try {
+
+		const modelsRes = await fetch( `https://generativelanguage.googleapis.com/v1beta/models?key=${geminiToken}` );
+		if ( ! modelsRes.ok ) {
+
+			console.error( 'Failed to fetch Gemini models. Please check your API token.' );
+			process.exit( 1 );
+
+		}
+
+		const modelsData = await modelsRes.json();
+		const rawModels = modelsData.models
+			.filter( m => m.name.includes( 'gemini' ) && m.supportedGenerationMethods && m.supportedGenerationMethods.includes( 'generateContent' ) )
+			.filter( m => ! m.name.includes( 'vision' ) )
+			.map( m => {
+
+				return {
+					id: m.name.replace( 'models/', '' ),
+					displayName: m.displayName || m.name.replace( 'models/', '' )
+				};
+
+			} );
+
+		// Filter out unwanted and noisy models
+		const filteredModels = rawModels.filter( m => {
+
+			const id = m.id.toLowerCase();
+			const name = m.displayName.toLowerCase();
+			if ( name.includes( 'nano' ) || name.includes( 'banana' ) || id.includes( 'nano' ) || id.includes( 'banana' ) ) return false;
+			if ( id.includes( '-latest' ) || id.includes( 'computer-use' ) || id.includes( 'robotics' ) || id.includes( '-tts' ) ) return false;
+			// Filter out very specific preview builds if they are too noisy, but we'll leave most previews as per the example.
+			return true;
+
+		} );
+
+		// Group by version
+		const groups = {};
+		filteredModels.forEach( m => {
+
+			// Try to extract version like "2.5", "3.1", "2.0", "3"
+			const versionMatch = m.id.match( /gemini-(\d+(?:\.\d+)?)/ );
+			if ( versionMatch ) {
+
+				const version = versionMatch[ 1 ];
+				if ( ! groups[ version ] ) groups[ version ] = [];
+				groups[ version ].push( m );
+
+			}
+
+		} );
+
+		const sortedVersions = Object.keys( groups ).sort( ( a, b ) => parseFloat( b ) - parseFloat( a ) );
+
+		if ( sortedVersions.length === 0 ) {
+
+			console.error( 'No suitable Gemini models found after filtering.' );
+			process.exit( 1 );
+
+		}
+
+		console.log( '\nSelect Gemini Version Family:' );
+		sortedVersions.forEach( ( v, i ) => {
+
+			console.log( `${i + 1}: Gemini ${v}` );
+
+		} );
+
+		const versionChoice = await askQuestion( `Enter choice (1-${sortedVersions.length}) [1]: ` );
+		let vIndex = parseInt( versionChoice, 10 );
+		if ( isNaN( vIndex ) || vIndex < 1 || vIndex > sortedVersions.length ) {
+
+			vIndex = 1;
+
+		}
+
+		const selectedVersion = sortedVersions[ vIndex - 1 ];
+		const familyModels = groups[ selectedVersion ].reverse(); // latest variants first
+
+		console.log( `\nSelect Model for Gemini ${selectedVersion}:` );
+		familyModels.forEach( ( m, i ) => {
+
+			console.log( `${i + 1}: ${m.displayName} (${m.id})` );
+
+		} );
+
+		const modelChoice = await askQuestion( `Enter choice (1-${familyModels.length}) [1]: ` );
+		let mIndex = parseInt( modelChoice, 10 );
+		if ( isNaN( mIndex ) || mIndex < 1 || mIndex > familyModels.length ) {
+
+			mIndex = 1;
+
+		}
+
+		return familyModels[ mIndex - 1 ].id;
+
+	} catch ( e ) {
+
+		console.error( 'Error fetching models:', e.message );
+		process.exit( 1 );
+
+	}
+
+}
+
+async function fetchMilestoneData( repo, milestoneNumber, headers ) {
+
+	const milestoneRes = await fetch( `https://api.github.com/repos/${repo}/milestones/${milestoneNumber}`, { headers } );
+	if ( ! milestoneRes.ok ) {
+
+		console.error( 'Failed to fetch milestone. Check if the milestone exists or rate limit exceeded.' );
+		process.exit( 1 );
+
+	}
+
+	const milestoneData = await milestoneRes.json();
+	return { title: milestoneData.title, closed_issues: milestoneData.closed_issues };
+
+}
+
+async function fetchAndParsePRs( repo, milestoneNumber, perPage, headers, totalExpectedPRs ) {
+
+	let page = 1;
+	let totalPages = '?';
+	let totalPRs = 0;
+	let processedPRs = 0;
+
+	let prDescriptionsForAI = '';
+
+	const barWidth = 40;
+
+	function updateProgress() {
+
+		if ( totalExpectedPRs > 0 ) {
+
+			const done = processedPRs;
+			const total = totalExpectedPRs;
+			const filled = Math.min( barWidth, Math.round( barWidth * done / total ) );
+			const bar = '█'.repeat( filled ) + '░'.repeat( Math.max( 0, barWidth - filled ) );
+			const pct = Math.min( 100, Math.round( 100 * done / total ) );
+			process.stderr.write( `\r  ${bar} ${pct}% (${done}/${total})` );
+
+		} else {
+
+			process.stderr.write( `\rFetching PR page ${page}/${totalPages}...` );
+
+		}
+
+	}
+
+	// We no longer need the initial 1-page fetch because we have totalExpectedPRs
+	while ( true ) {
+
+		if ( totalExpectedPRs === 0 ) {
+
+			process.stdout.write( `\rFetching PR page ${page}/${totalPages}...` );
+
+		}
+
+		let pageContent = '';
+		let isLastPage = false;
+
+		const res = await fetch( `https://api.github.com/repos/${repo}/issues?milestone=${milestoneNumber}&state=closed&per_page=${perPage}&page=${page}`, { headers } );
+		if ( ! res.ok ) {
+
+			console.error( '\nFailed to fetch PRs.' );
+			process.exit( 1 );
+
+		}
+
+		if ( totalPages === '?' || totalExpectedPRs === 0 ) {
+
+			const linkHeader = res.headers.get( 'link' );
+			if ( linkHeader ) {
+
+				const lastPageMatch = linkHeader.match( /page=(\d+)>; rel="last"/ );
+				if ( lastPageMatch ) {
+
+					totalPages = lastPageMatch[ 1 ];
+					if ( totalExpectedPRs === 0 ) process.stdout.write( `\rFetching PR page ${page}/${totalPages}...` );
+
+				}
+
+			}
+
+		}
+
+		const data = await res.json();
+		if ( data.length === 0 ) break;
+
+		const prsInData = data.filter( issue => issue.pull_request && issue.pull_request.merged_at );
+
+		for ( const pr of prsInData ) {
+
+			let title = pr.title;
+			let category = 'Others';
+
+			const catMatch = title.match( /^([^:]+):\s*(.*)$/ );
+			if ( catMatch ) {
+
+				category = catMatch[ 1 ].trim();
+				title = catMatch[ 2 ].trim();
+
+			}
+
+			title = title.replace( /\s*\(#\d+\)\s*$/, '' ).trim();
+			title = title.replace( /\.\s*$/, '' );
+
+			const authors = new Set();
+			if ( pr.user && pr.user.login ) authors.add( pr.user.login );
+
+			if ( pr.body ) {
+
+				const coAuthRegex1 = /Co-authored-by:\s*(?:@)?([a-zA-Z0-9-]+)/gi;
+				let m;
+				while ( ( m = coAuthRegex1.exec( pr.body ) ) !== null ) {
+
+					authors.add( m[ 1 ] );
+
+				}
+
+			}
+
+			const authorsList = Array.from( authors ).map( a => `@${a}` ).join( ', ' );
+
+			pageContent += `## ${category}: ${title} #${pr.number} (${authorsList})\n\n${pr.body || ''}\n\n`;
+
+			processedPRs ++;
+
+			if ( totalExpectedPRs > 0 ) updateProgress();
+
+		}
+
+		if ( data.length < perPage ) isLastPage = true;
+
+		prDescriptionsForAI += pageContent;
+
+		const prRegex = /^## (.*?):\s*(.*?)\s+#(\d+)\s+\((.*?)\)(?:\r?\n)([\s\S]*?)(?=\n## |$)/gm;
+		while ( prRegex.exec( pageContent ) !== null ) {
+
+			totalPRs ++;
+
+		}
+
+		if ( isLastPage ) break;
+
+		page ++;
+
+	}
+
+	if ( totalExpectedPRs > 0 ) {
+
+		const bar = '█'.repeat( barWidth );
+		process.stderr.write( `\r  ${bar} 100% (${totalPRs}/${totalPRs})` );
+		process.stderr.write( '\n\n' );
+
+	}
+
+	return { prDescriptionsForAI };
+
+}
+
+async function fetchAISummary( releaseNumber, prDescriptionsForAI, geminiModel, geminiToken ) {
+
+	console.error( `Generating AI Summary with Gemini (${geminiModel})...` );
+
+	const promptTemplatePath = path.join( __dirname, 'changelog.summary.md' );
+	const promptTemplate = fs.readFileSync( promptTemplatePath, 'utf8' );
+
+	const examplesChanges = getExamplesChanges( releaseNumber );
+
+	const aiPrompt = promptTemplate
+		.replace( '{RELEASE}', releaseNumber )
+		.replace( '{DESCRIPTION}', prDescriptionsForAI )
+		.replace( '{EXAMPLES_CHANGES}', examplesChanges );
+
+	try {
+
+		// codeql[js/file-access-to-http] - False Positive: The prompt template file is intentionally sent to the LLM API.
+		const geminiRes = await fetch( `https://generativelanguage.googleapis.com/v1beta/models/${geminiModel}:generateContent?key=${geminiToken}`, {
+			method: 'POST',
+			headers: { 'Content-Type': 'application/json' },
+			body: JSON.stringify( {
+				contents: [ { parts: [ { text: aiPrompt } ] } ]
+			} )
+		} );
+
+		if ( geminiRes.ok ) {
+
+			const geminiData = await geminiRes.json();
+			if ( geminiData.candidates && geminiData.candidates[ 0 ] && geminiData.candidates[ 0 ].content ) {
+
+				const aiText = geminiData.candidates[ 0 ].content.parts[ 0 ].text;
+				const aiFooter = `\n\n*Generated by ${geminiModel}*`;
+				return `## AI Summary (r${releaseNumber})\n\n${aiText}${aiFooter}`;
+
+			}
+
+		} else {
+
+			console.error( 'Failed to generate AI summary:', await geminiRes.text() );
+
+		}
+
+	} catch ( e ) {
+
+		console.error( 'Error generating AI summary:', e.message );
+
+	}
+
+	return '';
+
+}
+
+async function generateChangelog() {
+
+	const { releaseNumber, milestoneNumber, perPage } = getReleaseAndCacheArgs();
+
+	const githubToken = await askQuestion( 'Enter GitHub API Token (or press Enter to skip): ' );
+	const geminiToken = await askQuestion( 'Enter Gemini API Token (or press Enter to skip AI summary): ' );
+
+	let geminiModel = '';
+	if ( geminiToken ) {
+
+		geminiModel = await selectGeminiModel( geminiToken );
+
+	}
+
+	const repo = 'mrdoob/three.js';
+
+	console.error( `Fetching milestone for release r${releaseNumber} (ID: ${milestoneNumber})...` );
+
+	const headers = {
+		'User-Agent': 'NodeJS/ThreeJS-Changelog',
+		'Accept': 'application/vnd.github.v3+json'
+	};
+
+	if ( githubToken ) {
+
+		headers[ 'Authorization' ] = `token ${githubToken}`;
+
+	}
+
+	const milestoneData = await fetchMilestoneData( repo, milestoneNumber, headers );
+	const milestoneName = milestoneData.title;
+
+	console.error( `Fetching PRs for milestone: ${milestoneName}...` );
+
+	const { prDescriptionsForAI } = await fetchAndParsePRs( repo, milestoneNumber, perPage, headers, milestoneData.closed_issues );
+
+	let summary = '';
+
+	if ( geminiToken && prDescriptionsForAI ) {
+
+		summary = await fetchAISummary( releaseNumber, prDescriptionsForAI, geminiModel, geminiToken );
+
+	} else if ( prDescriptionsForAI ) {
+
+		summary = prDescriptionsForAI;
+
+	}
+
+	console.log( '\n' + '-'.repeat( 100 ) + '\n' );
+	console.log( summary );
+	console.log( '\n' );
+
+}
+
+generateChangelog().catch( console.error );
diff --git a/utils/changelog.summary.md b/utils/changelog.summary.md
new file mode 100644
index 00000000000000..afb07ceb39e901
--- /dev/null
+++ b/utils/changelog.summary.md
@@ -0,0 +1,20 @@
+You are an expert technical writer and developer advocate analyzing the changes of a new release of the Three.js library (release r{RELEASE}).
+Here are the descriptions of the Pull Requests merged in this release:
+
+{DESCRIPTION}
+
+And here is the list of examples added, modified, and removed in this release:
+{EXAMPLES_CHANGES}
+
+Please provide the following information formatted in Markdown (ALL EXPLANATIONS MUST BE WRITTEN IN ENGLISH):
+- A detailed and comprehensive overview of the most important changes, including new features, major refactorings, API additions/deprecations, and optimizations.
+- Group the changes logically by components (e.g., Core, WebGPU, Renderer, Materials, Geometries, Loaders, Nodes, Editor). You MUST use the exact heading level `##` for each component category (e.g., `## Core`, etc.).
+- For each significant change, provide technical context explaining *what* changed and *why* it matters to developers. Give detailed and complete descriptions, avoiding vague or generic statements.
+- IMPORTANT: Add an extra blank line between each bullet point in your lists for better readability.
+- A section for Examples. You MUST divide it into three sub-categories: `- **New Examples:**`, `- **Modified Examples:**`, and `- **Removed Examples:**` (if applicable). Under each sub-category, group related examples compactly into single bullet points. You MUST place the description on a new line purely using Markdown formatting (a regular line break and indentation), without HTML tags. For the modified ones, briefly explain what was changed. For new examples, briefly describe their purpose. Format each bullet EXACTLY like this:
+  - **[example1](https://...), [example2](https://...)**:
+  General description
+  Whenever you list an example, you MUST format it as a markdown link pointing to its URL on the Three.js website, exactly like this: `[example_name](https://threejs.org/examples/#example_name)`.
+- Migration tips for users upgrading to this new version. Use the exact heading "## Migration Tips" (without any numbering). Detail any breaking changes or required code updates. IMPORTANT: If a property or feature was created and then renamed or removed within this same release milestone, DO NOT include it as a migration tip.
+
+Output only the markdown content, without extra code block delimiters. Do not include a code examples section.