pi-free 2.0.11 → 2.0.12

package/CHANGELOG.md

@@ -5,7 +5,39 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## [2.0.11] - 2026-05-08
+## [Unreleased]
+
+## [2.0.12] - 2026-05-13
+
+### Added
+
+- **Novita AI provider** — OpenAI-compatible API at `api.novita.ai/openai/v1` with 100+ open-source models. Non-standard but rich metadata: per-model pricing (`input_token_price_per_m`), context size, max output tokens, reasoning/vision features, and model descriptions. 3 free models, 99 paid.
+
+- **FastRouter provider** — OpenRouter-compatible API at `api.fastrouter.ai/api/v1` with 170+ models. Always discovered (no auth needed for model listing). Full pricing, context lengths, and feature metadata. 129 text models (6 free, 123 paid) after filtering image/video. Set `FASTROUTER_API_KEY` for chat completions.
+
+- **Dynamic model fetching for OpenCode and OpenRouter** — Pi's built-in providers now get their models fetched dynamically from the API (`opencode.ai/zen/v1/models` and `openrouter.ai/api/v1/models`), same as Mistral, Groq, Cerebras, and xAI. Overwrites Pi's defaults with the full model list. OpenCode uses name-based free detection (API returns no pricing); OpenRouter uses full cost-based detection.
+
+- **API key reading from `~/.pi/agent/auth.json`** — `getOpencodeApiKey()` and `getOpenrouterApiKey()` now fall back to Pi's auth.json when the env var isn't set, matching how Pi's built-in providers read their keys.
+
+### Changed
+
+- **`_pricingKnown` guard in `isFreeModel`** — Providers can now signal whether pricing data is authoritative. When `_pricingKnown` is explicitly `false` (API returned no pricing), `isFreeModel` falls back to name-only detection (checks for "free" in the model name). This eliminates false positives where missing pricing data was treated as $0 cost. All affected providers (ZenMux, Together, CrofAI, dynamic-built-in, fetchOpenAICompatibleModels, deepinfra, sambanova, novita) now set this flag correctly.
+
+- **All providers now use `isFreeModel` consistently** — Together switched from hardcoded `cost===0` check to `isFreeModel`. DeepInfra and SambaNova switched from manual free lists to `isFreeModel` with proper `_pricingKnown` metadata. NVIDIA, Codestral, and Ollama explicitly documented as free-tier providers (`freeModels = allModels`).
+
+- **Unified OpenRouter-based providers** — Kilo, OpenRouter, and Cline now share the same `fetchOpenRouterCompatibleModels` / OpenRouter API logic.
+
+### Removed
+
+- **`DEFAULT_MIN_SIZE_B` (30B minimum model size filter)** — Removed from `model-fetcher.ts` and `cline-models.ts`. All models are now shown regardless of parameter count. NVIDIA still uses its own 70B threshold (`NVIDIA_MIN_SIZE_B`).
+
+### Fixed
+
+- **ZenMux false free classifications** — Models without `pricings` data (DeepSeek Chat V3.1, Kimi K2 0711, Claude 3.7 Sonnet) were incorrectly classified as free because missing pricing defaulted to $0. Fixed to 3 genuinely free models (down from 6 false positives).
+
+- **Together AI, CrofAI, dynamic-built-in missing-pricing false positives** — Same `?? 0` pattern across multiple providers could mark unpriced models as free. All now set `_pricingKnown: false` when pricing is absent from the API response.
+
+## [2.0.10] - 2026-05-08
 
 ### Fixed
 

package/README.md

@@ -64,6 +64,7 @@ Free models are shown by default — look for the provider prefixes:
 - `crofai/` — CrofAI OpenAI-compatible API (streaming, reasoning models)
 - `codestral/` — Codestral via Mistral (free Experiment plan: 2 req/min, 1B tokens/month)
 - `deepinfra/` — DeepInfra inference cloud ($5 one-time trial credit, no credit card)
+- `novita/` — Novita AI (100+ open-source models, OpenAI-compatible, 3 free models)
 
 > **Note:** Paid providers may occasionally offer free models or promotional credits. The `isFreeModel` helper automatically detects free models based on provider pricing data or model names containing "free". For providers that don't expose pricing (like CrofAI), only models with "free" in their names are marked as free.
 
@@ -74,6 +75,9 @@ Free models are shown by default — look for the provider prefixes:
 - `cerebras/` — Cerebras models (when `CEREBRAS_API_KEY` set)
 - `xai/` — xAI models (when `XAI_API_KEY` set)
 - `huggingface/` — Hugging Face models (when `HF_TOKEN` set)
+- `opencode/` — OpenCode models (fetched from opencode.ai/zen/v1, when `OPENCODE_API_KEY` set)
+- `openrouter/` — OpenRouter models (fetched from openrouter.ai, when `OPENROUTER_API_KEY` set)
+- `fastrouter/` — FastRouter models (always discovered, 170+ models, no auth for listing)
 
 **Note:** Fireworks is now a [built-in Pi provider](https://github.com/badlogic/pi-mono/blob/main/packages/coding-agent/CHANGELOG.md#0681---2026-04-22) — no extension needed. Set `FIREWORKS_API_KEY` to use it directly.
 
@@ -100,6 +104,8 @@ Want to see paid models too? Run the toggle command for your provider:
 /toggle-together  # Toggle Together AI (💳 trial credit provider)
 /toggle-sambanova # Toggle SambaNova (🔄 freemium)
 /toggle-llm7      # Toggle LLM7 (✅ free gateway)
+/toggle-novita    # Toggle Novita AI (💳 paid — 3 free models)
+/toggle-fastrouter # Toggle FastRouter (🔧 dynamic — always discovered)
 ```
 
 **Notes:**
@@ -442,28 +448,28 @@ export SAMBANOVA_API_KEY="..."
 
 Each provider has toggle commands to switch between free and all models:
 
-| Command               | Action                                                   |
-| --------------------- | -------------------------------------------------------- |
-| `/toggle-opencode`    | Toggle between free/all OpenCode models                  |
-| `/toggle-kilo`        | Toggle between free/all Kilo models                      |
-| `/toggle-openrouter`  | Toggle between free/all OpenRouter models                |
-| `/toggle-cline`       | Toggle between free/all Cline models                     |
-| `/toggle-nvidia`      | Toggle between free/all NVIDIA models                    |
-| `/toggle-ollama`      | Toggle between free/all Ollama Cloud models              |
-| `/toggle-mistral`     | Toggle between free/all Mistral models (🔧 dynamic)      |
-| `/toggle-groq`        | Toggle between free/all Groq models (🔧 dynamic)         |
-| `/toggle-cerebras`    | Toggle between free/all Cerebras models (🔧 dynamic)     |
-| `/toggle-xai`         | Toggle between free/all xAI models (🔧 dynamic)          |
-| `/toggle-huggingface` | Toggle between free/all Hugging Face models (🔧 dynamic) |
-| `/toggle-codestral`   | Toggle Codestral (💳 paid)                               |
-| `/toggle-deepinfra`   | Toggle DeepInfra (💳 trial credit)                       |
-| `/toggle-together`    | Toggle Together AI (💳 trial credit)                     |
-| `/toggle-sambanova`   | Toggle SambaNova (🔄 freemium)                           |
-| `/toggle-llm7`        | Toggle LLM7 (✅ free gateway)                            |
-| `/toggle-zenmux`      | Toggle ZenMux (💳 paid)                                  |
-| `/toggle-crofai`      | Toggle CrofAI (💳 paid)                                  |
-| `/ollama-cloud-refresh` | Re-fetch Ollama Cloud models live (no restart needed)  |
-| `/probe-ollama`       | Test Ollama Cloud models for 403 errors (auto-hide)      |
+| Command                 | Action                                                   |
+| ----------------------- | -------------------------------------------------------- |
+| `/toggle-opencode`      | Toggle between free/all OpenCode models                  |
+| `/toggle-kilo`          | Toggle between free/all Kilo models                      |
+| `/toggle-openrouter`    | Toggle between free/all OpenRouter models                |
+| `/toggle-cline`         | Toggle between free/all Cline models                     |
+| `/toggle-nvidia`        | Toggle between free/all NVIDIA models                    |
+| `/toggle-ollama`        | Toggle between free/all Ollama Cloud models              |
+| `/toggle-mistral`       | Toggle between free/all Mistral models (🔧 dynamic)      |
+| `/toggle-groq`          | Toggle between free/all Groq models (🔧 dynamic)         |
+| `/toggle-cerebras`      | Toggle between free/all Cerebras models (🔧 dynamic)     |
+| `/toggle-xai`           | Toggle between free/all xAI models (🔧 dynamic)          |
+| `/toggle-huggingface`   | Toggle between free/all Hugging Face models (🔧 dynamic) |
+| `/toggle-codestral`     | Toggle Codestral (💳 paid)                               |
+| `/toggle-deepinfra`     | Toggle DeepInfra (💳 trial credit)                       |
+| `/toggle-together`      | Toggle Together AI (💳 trial credit)                     |
+| `/toggle-sambanova`     | Toggle SambaNova (🔄 freemium)                           |
+| `/toggle-llm7`          | Toggle LLM7 (✅ free gateway)                            |
+| `/toggle-zenmux`        | Toggle ZenMux (💳 paid)                                  |
+| `/toggle-crofai`        | Toggle CrofAI (💳 paid)                                  |
+| `/ollama-cloud-refresh` | Re-fetch Ollama Cloud models live (no restart needed)    |
+| `/probe-ollama`         | Test Ollama Cloud models for 403 errors (auto-hide)      |
 
 **The toggle command:**
 

package/banner.svg

@@ -110,7 +110,7 @@
     <text x="20" y="70" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Ollama Cloud · ZenMux</text>
     <text x="20" y="90" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">CrofAI · Codestral · LLM7</text>
     <text x="20" y="110" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">DeepInfra · SambaNova</text>
-    <text x="20" y="130" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Together</text>
+    <text x="20" y="130" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Together · Novita · FastRouter</text>
   </g>
 
   <g transform="translate(970, 55)">

package/config.ts

@@ -32,6 +32,8 @@ interface PiFreeConfig {
 	deepinfra_api_key?: string;
 	sambanova_api_key?: string;
 	together_api_key?: string;
+	novita_api_key?: string;
+	fastrouter_api_key?: string;
 	kilo_free_only?: boolean;
 	hidden_models?: string[];
 	free_only?: boolean;
@@ -45,6 +47,8 @@ interface PiFreeConfig {
 	deepinfra_show_paid?: boolean;
 	sambanova_show_paid?: boolean;
 	together_show_paid?: boolean;
+	novita_show_paid?: boolean;
+	fastrouter_show_paid?: boolean;
 	openrouter_show_paid?: boolean;
 	opencode_show_paid?: boolean;
 }
@@ -59,6 +63,8 @@ const CONFIG_TEMPLATE: PiFreeConfig = {
 	deepinfra_api_key: "",
 	sambanova_api_key: "",
 	together_api_key: "",
+	novita_api_key: "",
+	fastrouter_api_key: "",
 
 	kilo_free_only: false,
 	hidden_models: [],
@@ -73,6 +79,8 @@ const CONFIG_TEMPLATE: PiFreeConfig = {
 	deepinfra_show_paid: false,
 	sambanova_show_paid: false,
 	together_show_paid: false,
+	novita_show_paid: false,
+	fastrouter_show_paid: false,
 	openrouter_show_paid: false,
 	opencode_show_paid: false,
 };
@@ -210,6 +218,17 @@ export function getTogetherShowPaid(): boolean {
 	return resolveBool("TOGETHER_SHOW_PAID", loadConfigFile().together_show_paid);
 }
 
+export function getNovitaShowPaid(): boolean {
+	return resolveBool("NOVITA_SHOW_PAID", loadConfigFile().novita_show_paid);
+}
+
+export function getFastrouterShowPaid(): boolean {
+	return resolveBool(
+		"FASTROUTER_SHOW_PAID",
+		loadConfigFile().fastrouter_show_paid,
+	);
+}
+
 export function getOllamaShowPaid(): boolean {
 	return resolveBool("OLLAMA_SHOW_PAID", loadConfigFile().ollama_show_paid);
 }
@@ -225,6 +244,41 @@ export function getOpencodeShowPaid(): boolean {
 	return resolveBool("OPENCODE_SHOW_PAID", loadConfigFile().opencode_show_paid);
 }
 
+export function getProviderShowPaid(providerId: string): boolean {
+	switch (providerId) {
+		case "kilo":
+			return getKiloShowPaid();
+		case "cline":
+			return getClineShowPaid();
+		case "zenmux":
+			return getZenmuxShowPaid();
+		case "crofai":
+			return getCrofaiShowPaid();
+		case "codestral":
+			return getCodestralShowPaid();
+		case "llm7":
+			return getLlm7ShowPaid();
+		case "deepinfra":
+			return getDeepinfraShowPaid();
+		case "sambanova":
+			return getSambanovaShowPaid();
+		case "together":
+			return getTogetherShowPaid();
+		case "novita":
+			return getNovitaShowPaid();
+		case "fastrouter":
+			return getFastrouterShowPaid();
+		case "ollama-cloud":
+			return getOllamaShowPaid();
+		case "openrouter":
+			return getOpenrouterShowPaid();
+		case "opencode":
+			return getOpencodeShowPaid();
+		default:
+			return false;
+	}
+}
+
 // =============================================================================
 // Global free-only mode
 // =============================================================================
@@ -273,6 +327,14 @@ export function getTogetherApiKey(): string | undefined {
 	return resolve("TOGETHER_AI_API_KEY", loadConfigFile().together_api_key);
 }
 
+export function getNovitaApiKey(): string | undefined {
+	return resolve("NOVITA_API_KEY", loadConfigFile().novita_api_key);
+}
+
+export function getFastrouterApiKey(): string | undefined {
+	return resolve("FASTROUTER_API_KEY", loadConfigFile().fastrouter_api_key);
+}
+
 export function getOllamaApiKey(): string | undefined {
 	return resolve("OLLAMA_API_KEY", loadConfigFile().ollama_api_key);
 }
@@ -302,12 +364,47 @@ export function getHfToken(): string | undefined {
 	return process.env.HF_TOKEN;
 }
 
+/**
+ * Read an API key from ~/.pi/agent/auth.json.
+ * Pi stores built-in provider keys there (opencode, openrouter, etc.).
+ * Falls back to env var if auth.json is missing or key not found.
+ */
+function readAuthJsonKey(
+	providerId: string,
+	envVar: string,
+): string | undefined {
+	// Check env var first (fast path)
+	const envVal = process.env[envVar];
+	if (envVal) return envVal;
+
+	// Check auth.json
+	try {
+		const authPath = join(PI_DIR, "agent", "auth.json");
+		if (!existsSync(authPath)) return undefined;
+		const raw = readFileSync(authPath, "utf8");
+		const auth = JSON.parse(raw) as Record<
+			string,
+			{ type?: string; key?: string }
+		>;
+		const entry = auth[providerId];
+		if (entry?.key?.trim()) return entry.key;
+	} catch {
+		// auth.json missing or corrupt — silently skip
+	}
+	return undefined;
+}
+
 /**
  * OpenRouter key — pi's built-in provider reads from ~/.pi/agent/auth.json.
- * pi-free only checks the env var to avoid stale keys from free.json.
+ * pi-free checks env var first, then auth.json.
  */
 export function getOpenrouterApiKey(): string | undefined {
-	return process.env.OPENROUTER_API_KEY;
+	return readAuthJsonKey("openrouter", "OPENROUTER_API_KEY");
+}
+
+/** OpenCode key — pi's built-in provider. Read from env or auth.json. */
+export function getOpencodeApiKey(): string | undefined {
+	return readAuthJsonKey("opencode", "OPENCODE_API_KEY");
 }
 
 // =============================================================================

package/constants.ts

@@ -22,6 +22,7 @@ export const PROVIDER_LLM7 = "llm7";
 export const PROVIDER_DEEPINFRA = "deepinfra";
 export const PROVIDER_SAMBANOVA = "sambanova";
 export const PROVIDER_TOGETHER = "together";
+export const PROVIDER_NOVITA = "novita";
 
 export const ALL_UNIQUE_PROVIDERS = [
 	PROVIDER_KILO,
@@ -38,6 +39,7 @@ export const ALL_UNIQUE_PROVIDERS = [
 	PROVIDER_DEEPINFRA,
 	PROVIDER_SAMBANOVA,
 	PROVIDER_TOGETHER,
+	PROVIDER_NOVITA,
 ] as const;
 
 // =============================================================================
@@ -59,6 +61,7 @@ export const BASE_URL_LLM7 = "https://api.llm7.io/v1";
 export const BASE_URL_DEEPINFRA = "https://api.deepinfra.com/v1/openai";
 export const BASE_URL_SAMBANOVA = "https://api.sambanova.ai/v1";
 export const BASE_URL_TOGETHER = "https://api.together.xyz/v1";
+export const BASE_URL_NOVITA = "https://api.novita.ai/openai/v1";
 
 /** Cline fetches free models from OpenRouter */
 export const BASE_URL_OPENROUTER = "https://openrouter.ai/api/v1";
@@ -84,7 +87,6 @@ export const CLINE_AUTH_TIMEOUT_MS = 5 * 60 * 1000; // 5 minutes
 // =============================================================================
 
 export const NVIDIA_MIN_SIZE_B = 70; // Minimum model size for NVIDIA NIM
-export const DEFAULT_MIN_SIZE_B = 30; // Default minimum model size for filtering
 
 // =============================================================================
 // Timeouts (milliseconds)

package/index.ts

@@ -40,6 +40,7 @@ import llm7 from "./providers/llm7/llm7.ts";
 import deepinfra from "./providers/deepinfra/deepinfra.ts";
 import sambanova from "./providers/sambanova/sambanova.ts";
 import together from "./providers/together/together.ts";
+import novita from "./providers/novita/novita.ts";
 import nvidia from "./providers/nvidia/nvidia.ts";
 import ollama from "./providers/ollama/ollama.ts";
 import zenmux from "./providers/zenmux/zenmux.ts";
@@ -57,7 +58,7 @@ function setupGlobalCommands(pi: ExtensionAPI) {
 		handler: async (_args, ctx) => {
 			const current = getGlobalFreeOnly();
 			const next = !current;
-			applyGlobalFilter(pi, next);
+			applyGlobalFilter(pi, next, { force: true });
 
 			const registry = getProviderRegistry();
 			const providerCount = registry.size;
@@ -210,10 +211,11 @@ export default async function piFreeEntry(pi: ExtensionAPI) {
 		deepinfra(pi),
 		sambanova(pi),
 		together(pi),
+		novita(pi),
 	]);
 
-	// Setup dynamic built-in providers (Mistral, Groq, Cerebras, xAI, Hugging Face)
-	// These only activate if the user has configured API keys (OpenRouter works without key too)
+	// Setup dynamic built-in providers (Mistral, Groq, Cerebras, xAI, Hugging Face,
+	// OpenRouter/OpenCode from Pi auth, and FastRouter public model discovery)
 	const { setupDynamicBuiltInProviders } = await import(
 		"./providers/dynamic-built-in/index.ts"
 	);

package/lib/built-in-toggle.ts

@@ -18,7 +18,11 @@ import type {
 } from "@earendil-works/pi-coding-agent";
 import { getOpencodeShowPaid, getOpenrouterShowPaid } from "../config.ts";
 import { createLogger } from "./logger.ts";
-import { isFreeModel, registerWithGlobalToggle } from "./registry.ts";
+import {
+	getProviderRegistry,
+	isFreeModel,
+	registerWithGlobalToggle,
+} from "./registry.ts";
 import { createToggleState } from "./toggle-state.ts";
 
 const _logger = createLogger("built-in-toggle");
@@ -55,68 +59,38 @@ let commandsRegistered = false;
 // =============================================================================
 
 export function setupBuiltInProviderToggles(pi: ExtensionAPI): void {
+	const activeConfigs = BUILT_IN_TOGGLE_PROVIDERS.filter(
+		(config) => !getProviderRegistry().has(config.id),
+	);
+
+	if (activeConfigs.length === 0) {
+		_logger.info(
+			"[built-in-toggle] OpenCode/OpenRouter already registered dynamically; skipping fallback capture",
+		);
+		return;
+	}
+
 	// Register toggle commands once (available even before models load)
 	if (!commandsRegistered) {
-		for (const config of BUILT_IN_TOGGLE_PROVIDERS) {
+		for (const config of activeConfigs) {
 			registerToggleCommand(pi, config);
 		}
-		setupStatusBar(pi);
+		setupStatusBar(pi, activeConfigs);
 		commandsRegistered = true;
 	}
 
 	// Capture built-in models on session start and apply initial filter
 	pi.on("session_start", async (_event, ctx) => {
-		const available = ctx.modelRegistry.getAvailable();
-
-		for (const config of BUILT_IN_TOGGLE_PROVIDERS) {
+		for (const config of activeConfigs) {
 			if (providerStates.has(config.id)) {
-				// Already captured this session — skip to avoid re-registering
+				// Already captured — skip to avoid re-registering
 				continue;
 			}
 
-			const providerModels = available.filter(
-				(m: Model<Api>) => m.provider === config.id,
-			);
-			if (providerModels.length === 0) continue;
-
-			const allModels = providerModels.map(modelToProviderConfig);
-			const freeModels = allModels.filter((m) =>
-				isFreeModel({ ...m, provider: config.id }, allModels),
-			);
-
-			const baseUrl = providerModels[0].baseUrl;
-			const api = providerModels[0].api;
-			const apiKeyEnv = getApiKeyEnvForProvider(config.id);
-
-			const reRegister = (models: ProviderModelConfig[]) => {
-				pi.registerProvider(config.id, {
-					baseUrl,
-					apiKey: apiKeyEnv,
-					api,
-					models,
-				});
-			};
-
-			const stored = { free: freeModels, all: allModels };
-			const toggleState = createToggleState<ProviderModelConfig>({
-				providerId: config.id,
-				initialShowPaid: config.getShowPaid(),
-				initialModels: stored,
-			});
-
-			providerStates.set(config.id, {
-				stored,
-				reRegister,
-				toggleState,
-			});
-
-			registerWithGlobalToggle(config.id, stored, reRegister, true);
+			const state = tryCaptureProvider(pi, config, ctx);
+			if (!state) continue;
 
-			_logger.info(
-				`[built-in-toggle] ${config.id}: captured ${allModels.length} models (${freeModels.length} free)`,
-			);
-
-			const applied = toggleState.applyCurrent(reRegister);
+			const applied = state.toggleState.applyCurrent(state.reRegister);
 			_logger.info(
 				`[built-in-toggle] ${config.id}: applied ${applied.mode} mode with ${applied.models.length} models`,
 			);
@@ -124,6 +98,58 @@ export function setupBuiltInProviderToggles(pi: ExtensionAPI): void {
 	});
 }
 
+// =============================================================================
+// On-demand model capture (called by toggle command when state is missing)
+// =============================================================================
+
+function tryCaptureProvider(
+	pi: ExtensionAPI,
+	config: BuiltInToggleConfig,
+	ctx: any,
+): BuiltInProviderState | undefined {
+	const available = ctx.modelRegistry.getAvailable();
+	const providerModels = available.filter(
+		(m: Model<Api>) => m.provider === config.id,
+	);
+	if (providerModels.length === 0) return undefined;
+
+	const allModels = providerModels.map(modelToProviderConfig);
+	const freeModels = allModels.filter((m: ProviderModelConfig) =>
+		isFreeModel({ ...m, provider: config.id }, allModels),
+	);
+
+	const baseUrl = providerModels[0].baseUrl;
+	const api = providerModels[0].api;
+	const apiKeyEnv = getApiKeyEnvForProvider(config.id);
+
+	const reRegister = (models: ProviderModelConfig[]) => {
+		pi.registerProvider(config.id, {
+			baseUrl,
+			apiKey: apiKeyEnv,
+			api,
+			models,
+		});
+	};
+
+	const stored = { free: freeModels, all: allModels };
+	const toggleState = createToggleState<ProviderModelConfig>({
+		providerId: config.id,
+		initialShowPaid: config.getShowPaid(),
+		initialModels: stored,
+	});
+
+	const state: BuiltInProviderState = { stored, reRegister, toggleState };
+	providerStates.set(config.id, state);
+
+	registerWithGlobalToggle(config.id, stored, reRegister, true);
+
+	_logger.info(
+		`[built-in-toggle] ${config.id}: captured ${allModels.length} models (${freeModels.length} free)`,
+	);
+
+	return state;
+}
+
 // =============================================================================
 // Per-provider toggle command
 // =============================================================================
@@ -136,13 +162,17 @@ function registerToggleCommand(
 	pi.registerCommand(commandName, {
 		description: `Toggle free/paid ${config.id} models`,
 		handler: async (_args, ctx) => {
-			const state = providerStates.get(config.id);
+			let state = providerStates.get(config.id);
 			if (!state) {
-				ctx.ui.notify(
-					`${config.id}: models not loaded yet. Start a session first.`,
-					"warning",
-				);
-				return;
+				// Models may have loaded after session_start — try on-demand capture
+				state = tryCaptureProvider(pi, config, ctx);
+				if (!state) {
+					ctx.ui.notify(
+						`${config.id}: models not loaded yet. Start a session first.`,
+						"warning",
+					);
+					return;
+				}
 			}
 
 			const applied = state.toggleState.toggle(state.reRegister);
@@ -185,12 +215,15 @@ function modelToProviderConfig(m: Model<Api>): ProviderModelConfig {
 // Status bar for provider selection
 // =============================================================================
 
-function setupStatusBar(pi: ExtensionAPI): void {
+function setupStatusBar(
+	pi: ExtensionAPI,
+	configs: BuiltInToggleConfig[],
+): void {
 	pi.on("model_select", (_event, ctx) => {
 		const selected = _event.model?.provider;
 
-		// Clear status for all built-in toggle providers
-		for (const config of BUILT_IN_TOGGLE_PROVIDERS) {
+		// Clear status for all fallback-captured built-in providers
+		for (const config of configs) {
 			if (selected !== config.id) {
 				ctx.ui.setStatus(`${config.id}-status`, undefined);
 			}

package/lib/registry.ts

@@ -9,7 +9,7 @@ import type {
 	ExtensionAPI,
 	ProviderModelConfig,
 } from "@earendil-works/pi-coding-agent";
-import { getFreeOnly, saveConfig } from "../config.ts";
+import { getFreeOnly, getProviderShowPaid, saveConfig } from "../config.ts";
 import { createLogger } from "./logger.ts";
 
 const _logger = createLogger("pi-free");
@@ -82,7 +82,7 @@ function detectPricingExposed(allModels: ProviderModelConfig[]): boolean {
  * @returns true if the model is definitively free per the provider's API
  */
 export function isFreeModel(
-	model: ProviderModelConfig & { provider?: string },
+	model: ProviderModelConfig & { provider?: string; _pricingKnown?: boolean },
 	allModels?: ProviderModelConfig[],
 ): boolean {
 	return isFreeModelInternal(model, allModels);
@@ -90,7 +90,7 @@ export function isFreeModel(
 
 // Internal implementation to work around TypeScript filter callback issues
 function isFreeModelInternal(
-	model: ProviderModelConfig & { provider?: string },
+	model: ProviderModelConfig & { provider?: string; _pricingKnown?: boolean },
 	allModels: ProviderModelConfig[] | undefined,
 ): boolean {
 	// Determine if pricing is exposed
@@ -106,12 +106,20 @@ function isFreeModelInternal(
 		pricingExposed = true;
 	}
 
-	// Route A: Pricing-exposed providers - use OR logic
-	// Model is free if EITHER cost is zero OR name contains "free"
+	// Route A: Pricing-exposed providers
+	// Model is free if EITHER cost is zero OR name contains "free".
+	// BUT: when _pricingKnown is explicitly false (API returned no pricing data),
+	// cost values are untrustworthy defaults — fall back to name-only detection.
 	if (pricingExposed) {
 		const isZeroCost =
 			(model.cost?.input ?? 0) === 0 && (model.cost?.output ?? 0) === 0;
 		const hasFreeInName = model.name.toLowerCase().includes("free");
+
+		// Pricing missing for this specific model — only trust name-based signal
+		if (model._pricingKnown === false) {
+			return hasFreeInName;
+		}
+
 		return isZeroCost || hasFreeInName;
 	}
 
@@ -156,12 +164,32 @@ export function getProviderRegistry(): ReadonlyMap<string, ProviderEntry> {
 // Global filter application
 // =============================================================================
 
+function showAllForProvider(providerId: string, entry: ProviderEntry): void {
+	const allModels =
+		entry.stored.all.length > 0 ? entry.stored.all : entry.stored.free;
+	if (allModels.length > 0) {
+		entry.reRegister(allModels);
+		_logger.info(
+			`[pi-free] ${providerId}: showing all ${allModels.length} models`,
+		);
+	}
+}
+
 function applyFilterToProvider(
 	providerId: string,
 	entry: ProviderEntry,
 	freeOnly: boolean,
+	force: boolean,
 ): void {
 	if (freeOnly) {
+		if (!force && getProviderShowPaid(providerId)) {
+			showAllForProvider(providerId, entry);
+			_logger.info(
+				`[pi-free] ${providerId}: preserved persisted all-models toggle`,
+			);
+			return;
+		}
+
 		if (entry.stored.free.length > 0) {
 			entry.reRegister(entry.stored.free);
 			_logger.info(
@@ -171,25 +199,21 @@ function applyFilterToProvider(
 			_logger.warn(`[pi-free] ${providerId}: no free models available`);
 		}
 	} else {
-		// Show all models (paid + free)
-		const allModels =
-			entry.stored.all.length > 0 ? entry.stored.all : entry.stored.free;
-		if (allModels.length > 0) {
-			entry.reRegister(allModels);
-			_logger.info(
-				`[pi-free] ${providerId}: showing all ${allModels.length} models`,
-			);
-		}
+		showAllForProvider(providerId, entry);
 	}
 }
 
-export function applyGlobalFilter(_pi: ExtensionAPI, freeOnly: boolean): void {
+export function applyGlobalFilter(
+	_pi: ExtensionAPI,
+	freeOnly: boolean,
+	options: { force?: boolean } = {},
+): void {
 	globalFreeOnly = freeOnly;
 	saveConfig({ free_only: freeOnly });
 
 	for (const [providerId, entry] of providerRegistry) {
 		try {
-			applyFilterToProvider(providerId, entry, freeOnly);
+			applyFilterToProvider(providerId, entry, freeOnly, options.force === true);
 		} catch (err) {
 			_logger.error(
 				`[pi-free] Failed to apply filter to ${providerId}`,