code-crusher · Sep 3, 2026 · Sep 2, 2026 · Sep 2, 2026 · Sep 2, 2026 · Sep 2, 2026
diff --git a/.changeset/axon-auto-default-model.md b/.changeset/axon-auto-default-model.md
diff --git a/.changeset/pasted-image-paths.md b/.changeset/pasted-image-paths.md
diff --git a/.changeset/remove-fireworks-provider.md b/.changeset/remove-fireworks-provider.md
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,5 +1,22 @@
 # Changelog
 
+## [v6.8.4] - 2026-09-03
+
+### Added
+
+- **Per-model usage visibility.** The usage dialog and a new Settings → Model Usage section now show each tracked OSS model's share of the shared plan pool as weekly/monthly percentages (no credit amounts), alongside the model's plan-cost multiplier badge (e.g. `5x cost`). The new settings section also renders the weekly/monthly plan windows with reset times and a refresh action, and is excluded from the save-button flow since it is read-only. Backend: `/axoncode/profile` now returns a `modelUsage` array (`model`, `multiplier`, `weeklyPercentage`, `monthlyPercentage`); charges for tracked OSS models are multiplied by their plan-cost multiplier before draining the shared pool.
+
+### Changed
+
+- **Dynamic model catalog synchronization.** Models are now dynamically fetched from the MatterAI backend `/v1/models` (or `/v1/web/models`) and registered into the client model registry (`registerDynamicKilocodeModels`), replacing static hardcoded lists while continuing to exclude deprecated axon models from active selection. Models automatically refresh on window focus, via the refresh button in the model selector, and through a 10-minute background poller. Cache bypass (`forceRefresh`) ensures database updates reflect immediately without restarting the extension.
+
+- **OSS model catalog replaces Axon models.** The KiloCode model catalog (extension `kilocode-models.ts` and the webview `useOpenRouterModelProviders` copy) now exposes seven OSS models — `meta/muse-spark-1.2-contributor` (Muse Spark 1.2 Contributor), `deepseek/deepseek-v4-flash-0731` (DeepSeek V4 Flash), `zai/glm-5.3` (GLM 5.3), `zai/glm-5.3-flash` (GLM 5.3 Flash), `gpt-5.6-sol` (GPT-5.6 Sol), `gpt-5.6-luna` (GPT-5.6 Luna), and `gemini-3.7-flash` (Gemini 3.7 Flash) — in place of the Axon context-window variants. Each OSS model carries its published per-token pricing (Muse Spark 1.2 Contributor $0.10/M input, $0.002/M cache read, $0.20/M output; DeepSeek V4 Flash $0.14/M input, $0.028/M cache read, $0.28/M output; GLM 5.3 $1.40/M input, $0.14/M cache read, $4.40/M output; GLM 5.3 Flash $0.15/M input, $0.03/M cache read, $0.50/M output; GPT-5.6 Sol $5/M input, $0.50/M cache read, $30/M output; GPT-5.6 Luna $0.20/M input, $0.02/M cache read, $1.20/M output; Gemini 3.7 Flash $0.75/M input, $0.075/M cache read, $3.75/M output). The default model is `deepseek/deepseek-v4-flash-0731` (`openRouterDefaultModelId` in `@roo-code/types`, plus the CLI `kilocodeModel` defaults and web-evals `MODEL_DEFAULT`). Stored Axon model selections are now stale and reset to the default on next launch via the existing `isValidKilocodeModel` stale-model check.
+
+### Fixed
+
+- **OpenRouter models no longer wiped on background refresh.** `refreshKilocodeModels` (window-focus refresh, 10-minute poller, startup fetch) previously posted `openrouter: {}` to the webview, blanking the OpenRouter model list until the next manual reload. It now fetches both `openrouter` and `kilocode-openrouter` in parallel (mirroring the `requestRouterModels` handler) and posts the merged payload, logging and skipping only the provider that failed.
+- **Stale model selections reset on launch.** The dynamic catalog is now fetched once at provider startup (after `ContextProxy` initialization) and the webview state re-posted, so the `isValidKilocodeModel` stale-model check runs against a populated catalog instead of the empty-catalog bypass. Previously, a stored axon model survived until the first focus- or webview-triggered fetch completed.
+
 ## [v6.8.2] - 2026-08-28
 
 ### Added

diff --git a/README.md b/README.md
@@ -85,13 +85,17 @@ Orbital provides a comprehensive suite of intelligent tools:
 
 ## 🤖 AI Models
 
-Choose the right model for your workflow — switch seamlessly between efficiency and power:
-
-| Model             | Credit Usage | Use Case              | Capabilities                                                                |
-| ----------------- | ------------ | --------------------- | --------------------------------------------------------------------------- |
-| **`axon-mini`**   | 0.5x         | Quick tasks           | Lightweight, fast, cost-effective for high-volume usage                     |
-| **`axon-code`**   | 0.8x         | Daily coding tasks    | High intelligence, balanced performance, best for regular coding            |
-| **`axon-code-2`** | 1x           | Complex agentic tasks | Maximum intelligence, complex context harness, best for complex development |
+Choose the right model for your workflow — switch seamlessly between efficiency and power. The model catalog is served dynamically from the MatterAI backend, so new models appear automatically in the model selector:
+
+| Model                          | Provider | Credit Usage | Use Case              | Capabilities                                                    |
+| ------------------------------ | -------- | ------------ | --------------------- | --------------------------------------------------------------- |
+| **Muse Spark 1.3 Contributor** | Meta     | 2x           | Everyday coding       | Open general-purpose model for everyday coding tasks            |
+| **DeepSeek V4 Flash**          | DeepSeek | 5x           | Fast, low-cost coding | Fast, low-cost open model for day-to-day coding tasks           |
+| **GLM 5.3**                    | Z.ai     | 4x           | Complex agentic tasks | Frontier open model for complex coding and long-running agents  |
+| **GLM 5.3 Flash**              | Z.ai     | 4x           | Everyday coding       | Fast, low-cost open model for everyday coding tasks             |
+| **GPT-5.6 Luna**               | OpenAI   | 2x           | Everyday coding       | Fast, low-cost open model for everyday coding tasks             |
+| **GPT-5.6 Sol**                | OpenAI   | 5x           | Complex reasoning     | Open reasoning model for complex coding and long-running agents |
+| **Gemini 3.8 Flash**           | Google   | 3x           | Everyday coding       | Fast, low-cost model for everyday coding tasks                  |
 
 ## 📦 Installation
 

diff --git a/apps/web-evals/src/lib/schemas.ts b/apps/web-evals/src/lib/schemas.ts
@@ -6,7 +6,7 @@ import { rooCodeSettingsSchema } from "@roo-code/types"
  * CreateRun
  */
 
-export const MODEL_DEFAULT = "axon-auto-232k"
+export const MODEL_DEFAULT = "deepseek/deepseek-v4-flash-0731"
 
 export const CONCURRENCY_MIN = 1
 export const CONCURRENCY_MAX = 25

diff --git a/cli/src/config/__tests__/persistence-provider-merge.test.ts b/cli/src/config/__tests__/persistence-provider-merge.test.ts
@@ -51,7 +51,7 @@ describe("Provider Merging", () => {
 		expect(result.config.providers[0].id).toBe("default")
 		expect(result.config.providers[0]).toHaveProperty("kilocodeToken")
 		expect(result.config.providers[0]).toHaveProperty("kilocodeModel")
-		expect(result.config.providers[0].kilocodeModel).toBe("axon-auto-232k")
+		expect(result.config.providers[0].kilocodeModel).toBe("deepseek/deepseek-v4-flash-0731")
 		expect(result.validation.valid).toBe(true)
 	})
 

diff --git a/cli/src/config/defaults.ts b/cli/src/config/defaults.ts
@@ -55,7 +55,7 @@ export const DEFAULT_CONFIG = {
 			id: "default",
 			provider: "kilocode",
 			kilocodeToken: "",
-			kilocodeModel: "axon-auto-232k",
+			kilocodeModel: "deepseek/deepseek-v4-flash-0731",
 		},
 	],
 	autoApproval: DEFAULT_AUTO_APPROVAL,

diff --git a/cli/src/constants/providers/models.ts b/cli/src/constants/providers/models.ts
@@ -475,7 +475,7 @@ export function sortModelsByPreference(models: ModelRecord): string[] {
 	}
 
 	// Sort rest alphabetically
-	restModelIds.sort((a, b) => a.localeCompare(b))
+	// restModelIds.sort((a, b) => a.localeCompare(b))
 
 	return [...preferredModelIds, ...restModelIds]
 }

diff --git a/cli/src/constants/providers/settings.ts b/cli/src/constants/providers/settings.ts
@@ -551,7 +551,7 @@ export const getProviderSettings = (provider: ProviderName, config: ProviderSett
 			return [
 				createFieldConfig("kilocodeToken", config),
 				createFieldConfig("kilocodeOrganizationId", config, "personal"),
-				createFieldConfig("kilocodeModel", config, "axon-auto-232k"),
+				createFieldConfig("kilocodeModel", config, "deepseek/deepseek-v4-flash-0731"),
 			]
 
 		default:
@@ -563,7 +563,7 @@ export const getProviderSettings = (provider: ProviderName, config: ProviderSett
  * Provider-specific default models
  */
 export const PROVIDER_DEFAULT_MODELS: Record<ProviderName, string> = {
-	kilocode: "axon-auto-232k",
+	kilocode: "deepseek/deepseek-v4-flash-0731",
 	anthropic: "claude-3-5-sonnet-20241022",
 	"openai-native": "gpt-4o",
 	openrouter: "anthropic/claude-3-5-sonnet",

diff --git a/cli/src/utils/browserAuth.ts b/cli/src/utils/browserAuth.ts
@@ -203,7 +203,7 @@ export async function performBrowserAuth(source: string = "axon-code-cli"): Prom
 					id: "default",
 					provider: "kilocode",
 					kilocodeToken: token,
-					kilocodeModel: "axon-auto-232k",
+					kilocodeModel: "deepseek/deepseek-v4-flash-0731",
 				},
 			],
 		}

diff --git a/packages/types/src/model.ts b/packages/types/src/model.ts
@@ -79,6 +79,9 @@ export const modelInfoSchema = z.object({
 	// forked_change start
 	displayName: z.string().nullish(),
 	preferredIndex: z.number().nullish(),
+	// Provider logo URL (SVG) from the MatterAI catalog; rendered by the UI
+	// on a white circular background.
+	iconUrl: z.string().nullish(),
 	// forked_change end
 	// Flag to indicate if the model is deprecated and should not be used
 	deprecated: z.boolean().optional(),

diff --git a/packages/types/src/providers/openrouter.ts b/packages/types/src/providers/openrouter.ts
@@ -1,20 +1,19 @@
 import type { ModelInfo } from "../model.js"
 
-// https://openrouter.ai/models?order=newest&supported_parameters=tools
-export const openRouterDefaultModelId = "axon-auto-232k"
+// MatterAI OSS models served through the MatterAI gateway
+export const openRouterDefaultModelId = "deepseek/deepseek-v4-flash-0731"
 
 export const openRouterDefaultModelInfo: ModelInfo = {
 	maxTokens: 64000,
 	contextWindow: 232_000,
 	supportsImages: true,
 	supportsComputerUse: false,
-	supportsPromptCache: false,
-	inputPrice: 1.0,
-	outputPrice: 4.0,
-	cacheWritesPrice: 0.0,
-	cacheReadsPrice: 0.0,
-	description:
-		"Axon Auto starts with Eido 3 Code Flash and dynamically selects Flash, Mini, or Pro as the task evolves. Pricing is dynamic and follows the model used for each request.",
+	supportsPromptCache: true,
+	inputPrice: 0.00000015,
+	outputPrice: 0.0000005,
+	cacheWritesPrice: 0,
+	cacheReadsPrice: 0.00000003,
+	description: "GLM 5.3 Flash is a fast, low cost open model for everyday coding tasks.",
 }
 
 export const OPENROUTER_DEFAULT_PROVIDER_NAME = "[default]"