fix(vscode-lm): simplify image support and exclude non-tool-call models

- Simplify IMAGE_CAPABLE_MODEL_PREFIXES to ['gpt', 'claude', 'gemini', 'o1', 'o3']
- Use startsWith prefix matching on model ID for cleaner logic
- Add claude-opus-41 to VSCODE_LM_STATIC_BLACKLIST (no tool_call support)
- Return text placeholder for URL images instead of empty Uint8Array
- Update tests to match new prefix-matching semantics

Source: https://models.dev/api.json (github-copilot provider models)

Author: hannesrudolph

src/api/providers/__tests__/vscode-lm.spec.ts

@@ -1,4 +1,5 @@
 import type { Mock } from "vitest"
+import { checkModelSupportsImages, IMAGE_CAPABLE_MODEL_PREFIXES } from "../vscode-lm"
 
 // Mocks must come first, before imports
 vi.mock("vscode", () => {
@@ -537,3 +538,92 @@ describe("VsCodeLmHandler", () => {
 		})
 	})
 })
+
+describe("checkModelSupportsImages", () => {
+	describe("OpenAI GPT models", () => {
+		it("should return true for all gpt-* models (GitHub Copilot)", () => {
+			// All GPT models in GitHub Copilot support images
+			expect(checkModelSupportsImages("gpt", "gpt-4o")).toBe(true)
+			expect(checkModelSupportsImages("gpt", "gpt-4.1")).toBe(true)
+			expect(checkModelSupportsImages("gpt", "gpt-5")).toBe(true)
+			expect(checkModelSupportsImages("gpt", "gpt-5.1")).toBe(true)
+			expect(checkModelSupportsImages("gpt", "gpt-5.2")).toBe(true)
+			expect(checkModelSupportsImages("gpt-mini", "gpt-5-mini")).toBe(true)
+			expect(checkModelSupportsImages("gpt-codex", "gpt-5.1-codex")).toBe(true)
+			expect(checkModelSupportsImages("gpt-codex", "gpt-5.2-codex")).toBe(true)
+			expect(checkModelSupportsImages("gpt-codex", "gpt-5.1-codex-max")).toBe(true)
+			expect(checkModelSupportsImages("gpt-codex", "gpt-5.1-codex-mini")).toBe(true)
+		})
+
+		it("should return true for o1 and o3 reasoning models", () => {
+			expect(checkModelSupportsImages("o1", "o1-preview")).toBe(true)
+			expect(checkModelSupportsImages("o1", "o1-mini")).toBe(true)
+			expect(checkModelSupportsImages("o3", "o3")).toBe(true)
+		})
+	})
+
+	describe("Anthropic Claude models", () => {
+		it("should return true for all claude-* models (GitHub Copilot)", () => {
+			// All Claude models in GitHub Copilot support images
+			expect(checkModelSupportsImages("claude-haiku", "claude-haiku-4.5")).toBe(true)
+			expect(checkModelSupportsImages("claude-opus", "claude-opus-4.5")).toBe(true)
+			expect(checkModelSupportsImages("claude-sonnet", "claude-sonnet-4")).toBe(true)
+			expect(checkModelSupportsImages("claude-sonnet", "claude-sonnet-4.5")).toBe(true)
+		})
+	})
+
+	describe("Google Gemini models", () => {
+		it("should return true for all gemini-* models (GitHub Copilot)", () => {
+			// All Gemini models in GitHub Copilot support images
+			expect(checkModelSupportsImages("gemini-pro", "gemini-2.5-pro")).toBe(true)
+			expect(checkModelSupportsImages("gemini-flash", "gemini-3-flash-preview")).toBe(true)
+			expect(checkModelSupportsImages("gemini-pro", "gemini-3-pro-preview")).toBe(true)
+		})
+	})
+
+	describe("non-vision models", () => {
+		it("should return false for grok models (text-only in GitHub Copilot)", () => {
+			// Grok is the only model family in GitHub Copilot that doesn't support images
+			expect(checkModelSupportsImages("grok", "grok-code-fast-1")).toBe(false)
+		})
+
+		it("should return false for models with non-matching prefixes", () => {
+			// Models that don't start with gpt, claude, gemini, o1, or o3
+			expect(checkModelSupportsImages("mistral", "mistral-large")).toBe(false)
+			expect(checkModelSupportsImages("llama", "llama-3-70b")).toBe(false)
+			expect(checkModelSupportsImages("unknown", "some-random-model")).toBe(false)
+		})
+	})
+
+	describe("case insensitivity", () => {
+		it("should match regardless of case", () => {
+			expect(checkModelSupportsImages("GPT", "GPT-4O")).toBe(true)
+			expect(checkModelSupportsImages("CLAUDE", "CLAUDE-SONNET-4")).toBe(true)
+			expect(checkModelSupportsImages("GEMINI", "GEMINI-2.5-PRO")).toBe(true)
+		})
+	})
+
+	describe("prefix matching", () => {
+		it("should only match IDs that start with known prefixes", () => {
+			// ID must START with the prefix, not just contain it
+			expect(checkModelSupportsImages("custom", "gpt-4o")).toBe(true) // ID starts with gpt
+			expect(checkModelSupportsImages("custom", "my-gpt-model")).toBe(false) // gpt not at start
+			expect(checkModelSupportsImages("custom", "not-claude-model")).toBe(false) // claude not at start
+		})
+	})
+})
+
+describe("IMAGE_CAPABLE_MODEL_PREFIXES", () => {
+	it("should export the model prefixes array", () => {
+		expect(Array.isArray(IMAGE_CAPABLE_MODEL_PREFIXES)).toBe(true)
+		expect(IMAGE_CAPABLE_MODEL_PREFIXES.length).toBeGreaterThan(0)
+	})
+
+	it("should include key model prefixes", () => {
+		expect(IMAGE_CAPABLE_MODEL_PREFIXES).toContain("gpt")
+		expect(IMAGE_CAPABLE_MODEL_PREFIXES).toContain("claude")
+		expect(IMAGE_CAPABLE_MODEL_PREFIXES).toContain("gemini")
+		expect(IMAGE_CAPABLE_MODEL_PREFIXES).toContain("o1")
+		expect(IMAGE_CAPABLE_MODEL_PREFIXES).toContain("o3")
+	})
+})

src/api/providers/vscode-lm.ts

@@ -591,46 +591,42 @@ export class VsCodeLmHandler extends BaseProvider implements SingleCompletionHan
 }
 
 /**
- * Model families known to support image inputs via VS Code Language Model API.
+ * Model ID prefixes that support image inputs via VS Code Language Model API.
  * These models support the LanguageModelDataPart.image() API introduced in VS Code 1.106+.
+ *
+ * All GitHub Copilot models with these prefixes support images.
+ * Only grok-* models don't support images (text only).
+ *
+ * Source: https://models.dev/api.json (github-copilot provider models)
  */
-const IMAGE_CAPABLE_MODEL_FAMILIES = [
-	// OpenAI models with vision capabilities
-	"gpt-4o",
-	"gpt-4o-mini",
-	"gpt-4.1",
-	"gpt-5",
-	"gpt-5-mini",
-	// Anthropic Claude models with vision
-	"claude-3.5-sonnet",
-	"claude-3-5-sonnet",
-	"claude-sonnet-4",
-	"claude-4-sonnet",
-	// Google Gemini models with vision
-	"gemini-2.0-flash",
-	"gemini-2.5-pro",
-	"gemini-pro-vision",
+export const IMAGE_CAPABLE_MODEL_PREFIXES = [
+	"gpt", // All GPT models (gpt-4o, gpt-4.1, gpt-5, gpt-5.1, gpt-5.2, gpt-5-mini, gpt-5.1-codex, etc.)
+	"claude", // All Claude models (claude-haiku-4.5, claude-opus-4.5, claude-sonnet-4, claude-sonnet-4.5)
+	"gemini", // All Gemini models (gemini-2.5-pro, gemini-3-flash-preview, gemini-3-pro-preview)
+	"o1", // OpenAI o1 reasoning models
+	"o3", // OpenAI o3 reasoning models
 ]
 
 /**
- * Checks if a model supports image inputs based on its family or ID.
- * @param family The model family (e.g., "gpt-4o", "claude-3.5-sonnet")
+ * Checks if a model supports image inputs based on its model ID.
+ * Uses prefix matching against known image-capable model families.
+ *
+ * @param _family The model family (unused, kept for API compatibility)
  * @param id The model ID
  * @returns true if the model supports image inputs
  */
-function checkModelSupportsImages(family: string, id: string): boolean {
-	// Check if the family matches any known image-capable model
-	const familyLower = family.toLowerCase()
+export function checkModelSupportsImages(_family: string, id: string): boolean {
 	const idLower = id.toLowerCase()
-
-	return IMAGE_CAPABLE_MODEL_FAMILIES.some(
-		(capableFamily) =>
-			familyLower.includes(capableFamily.toLowerCase()) || idLower.includes(capableFamily.toLowerCase()),
-	)
+	return IMAGE_CAPABLE_MODEL_PREFIXES.some((prefix) => idLower.startsWith(prefix))
 }
 
-// Static blacklist of VS Code Language Model IDs that should be excluded from the model list e.g. because they will never work
-const VSCODE_LM_STATIC_BLACKLIST: string[] = ["claude-3.7-sonnet", "claude-3.7-sonnet-thought"]
+// Static blacklist of VS Code Language Model IDs that should be excluded from the model list
+// e.g. because they don't support native tool calling or will never work
+const VSCODE_LM_STATIC_BLACKLIST: string[] = [
+	"claude-3.7-sonnet",
+	"claude-3.7-sonnet-thought",
+	"claude-opus-41", // Does not support native tool calling
+]
 
 export async function getVsCodeLmModels() {
 	try {

src/api/transform/__tests__/vscode-lm-format.spec.ts

@@ -256,6 +256,68 @@ describe("convertToVsCodeLmMessages", () => {
 		expect(imagePart.type).toBe("data")
 		expect(imagePart.mimeType).toBe("image/jpeg")
 	})
+
+	it("should return text placeholder for URL-based images", () => {
+		const messages: Anthropic.Messages.MessageParam[] = [
+			{
+				role: "user",
+				content: [
+					{ type: "text", text: "Check this image:" },
+					{
+						type: "image",
+						source: {
+							type: "url",
+							url: "https://example.com/image.png",
+						} as any,
+					},
+				],
+			},
+		]
+
+		const result = convertToVsCodeLmMessages(messages)
+
+		expect(result).toHaveLength(1)
+		expect(result[0].content).toHaveLength(2)
+
+		// First part should be text
+		const textPart = result[0].content[0] as MockLanguageModelTextPart
+		expect(textPart.type).toBe("text")
+		expect(textPart.value).toBe("Check this image:")
+
+		// Second part should be a text placeholder (not an empty DataPart)
+		const imagePlaceholder = result[0].content[1] as MockLanguageModelTextPart
+		expect(imagePlaceholder.type).toBe("text")
+		expect(imagePlaceholder.value).toContain("URL not supported")
+		expect(imagePlaceholder.value).toContain("https://example.com/image.png")
+	})
+
+	it("should return text placeholder for unknown image source types", () => {
+		const messages: Anthropic.Messages.MessageParam[] = [
+			{
+				role: "user",
+				content: [
+					{
+						type: "image",
+						source: {
+							type: "unknown",
+							media_type: "image/png",
+							data: "", // Required by type but ignored for unknown source types
+						} as any,
+					},
+				],
+			},
+		]
+
+		const result = convertToVsCodeLmMessages(messages)
+
+		expect(result).toHaveLength(1)
+		expect(result[0].content).toHaveLength(1)
+
+		// Should return a text placeholder for unknown source types
+		const placeholder = result[0].content[0] as MockLanguageModelTextPart
+		expect(placeholder.type).toBe("text")
+		expect(placeholder.value).toContain("unsupported source type")
+	})
 })
 
 describe("convertToAnthropicRole", () => {

src/api/transform/vscode-lm-format.ts

@@ -29,12 +29,14 @@ function asObjectSafe(value: any): object {
 }
 
 /**
- * Converts an Anthropic image block to a VS Code LanguageModelDataPart.
+ * Converts an Anthropic image block to a VS Code LanguageModelDataPart or TextPart.
  * Uses the new LanguageModelDataPart.image() API available in VS Code 1.106+.
  * @param imageBlock The Anthropic image block param
- * @returns A LanguageModelDataPart for the image
+ * @returns A LanguageModelDataPart for the image, or TextPart if the image cannot be converted
  */
-function convertImageToDataPart(imageBlock: Anthropic.ImageBlockParam): vscode.LanguageModelDataPart {
+function convertImageToDataPart(
+	imageBlock: Anthropic.ImageBlockParam,
+): vscode.LanguageModelDataPart | vscode.LanguageModelTextPart {
 	const source = imageBlock.source
 	const mediaType = source.media_type || "image/png"
 
@@ -47,18 +49,23 @@ function convertImageToDataPart(imageBlock: Anthropic.ImageBlockParam): vscode.L
 		}
 		return vscode.LanguageModelDataPart.image(bytes, mediaType)
 	} else if (source.type === "url") {
-		// For URL-based images, we create a placeholder since LanguageModelDataPart.image
-		// expects binary data. The URL would need to be fetched first.
-		// This is a limitation - URL images should be fetched and converted to base64 upstream.
+		// URL-based images cannot be directly converted - return a text placeholder
+		// explaining the limitation. URL images should be fetched and converted to base64 upstream.
 		console.warn(
-			"Roo Code <Language Model API>: URL-based images require fetching the image data first. Using placeholder.",
+			"Roo Code <Language Model API>: URL-based images are not supported by the VS Code LM API. " +
+				"Images must be provided as base64 data.",
+		)
+		return new vscode.LanguageModelTextPart(
+			`[Image from URL not supported: ${(source as any).url || "unknown URL"}. ` +
+				`VS Code LM API requires base64-encoded image data.]`,
 		)
-		return new vscode.LanguageModelDataPart(new Uint8Array(0), mediaType)
 	}
 
-	// Fallback for unknown source types
+	// Fallback for unknown source types - return a text placeholder
 	console.warn(`Roo Code <Language Model API>: Unknown image source type: ${(source as any).type}`)
-	return new vscode.LanguageModelDataPart(new Uint8Array(0), mediaType)
+	return new vscode.LanguageModelTextPart(
+		`[Image with unsupported source type "${(source as any).type}" cannot be displayed]`,
+	)
 }
 
 export function convertToVsCodeLmMessages(