diff --git a/packages/opencode/src/provider/litellm.ts b/packages/opencode/src/provider/litellm.ts
new file mode 100644
index 00000000000..bf5431aeed6
--- /dev/null
+++ b/packages/opencode/src/provider/litellm.ts
@@ -0,0 +1,170 @@
+import { Log } from "../util/log"
+import { Env } from "../env"
+import type { Provider } from "./provider"
+
+export namespace LiteLLM {
+  const log = Log.create({ service: "litellm" })
+
+  interface ModelInfoEntry {
+    model_name: string
+    litellm_params?: {
+      model?: string
+      [key: string]: unknown
+    }
+    model_info?: {
+      id?: string
+      input_cost_per_token?: number
+      output_cost_per_token?: number
+      cache_read_input_token_cost?: number
+      cache_creation_input_token_cost?: number
+      max_tokens?: number
+      max_input_tokens?: number
+      max_output_tokens?: number
+      supports_function_calling?: boolean
+      supports_vision?: boolean
+      supports_pdf_input?: boolean
+      supports_audio_input?: boolean
+      supports_audio_output?: boolean
+      supports_video_input?: boolean
+      supports_prompt_caching?: boolean
+      supports_reasoning?: boolean
+      [key: string]: unknown
+    }
+  }
+
+  const INTERLEAVED_MODELS = ["claude", "anthropic"]
+
+  function isWildcard(name: string): boolean {
+    return name.includes("*") || name.includes("/*")
+  }
+
+  function inferInterleaved(
+    underlyingModel: string | undefined,
+  ): Provider.Model["capabilities"]["interleaved"] {
+    if (!underlyingModel) return false
+    const lower = underlyingModel.toLowerCase()
+    if (INTERLEAVED_MODELS.some((m) => lower.includes(m))) return true
+    return false
+  }
+
+  function toModel(entry: ModelInfoEntry): Provider.Model | undefined {
+    if (isWildcard(entry.model_name)) return undefined
+
+    const info = entry.model_info ?? {}
+    const underlyingModel = entry.litellm_params?.model
+
+    const inputCost = (info.input_cost_per_token ?? 0) * 1_000_000
+    const outputCost = (info.output_cost_per_token ?? 0) * 1_000_000
+    const cacheReadCost = (info.cache_read_input_token_cost ?? 0) * 1_000_000
+    const cacheWriteCost = (info.cache_creation_input_token_cost ?? 0) * 1_000_000
+
+    return {
+      id: entry.model_name,
+      providerID: "litellm",
+      name: entry.model_name,
+      api: {
+        id: entry.model_name,
+        url: "",
+        npm: "@ai-sdk/openai-compatible",
+      },
+      status: "active",
+      headers: {},
+      options: {},
+      cost: {
+        input: inputCost,
+        output: outputCost,
+        cache: {
+          read: cacheReadCost,
+          write: cacheWriteCost,
+        },
+      },
+      limit: {
+        context: (info.max_tokens ?? info.max_input_tokens ?? 128_000) as number,
+        output: (info.max_output_tokens ?? 8_192) as number,
+      },
+      capabilities: {
+        temperature: true,
+        reasoning: info.supports_reasoning ?? false,
+        attachment: (info.supports_vision || info.supports_pdf_input) ?? false,
+        toolcall: info.supports_function_calling ?? true,
+        input: {
+          text: true,
+          audio: info.supports_audio_input ?? false,
+          image: info.supports_vision ?? false,
+          video: info.supports_video_input ?? false,
+          pdf: info.supports_pdf_input ?? false,
+        },
+        output: {
+          text: true,
+          audio: info.supports_audio_output ?? false,
+          image: false,
+          video: false,
+          pdf: false,
+        },
+        interleaved: inferInterleaved(underlyingModel),
+      },
+      release_date: "",
+      variants: {},
+    }
+  }
+
+  export async function discover(
+    host: string,
+    options?: {
+      apiKey?: string
+      headers?: Record<string, string>
+      timeout?: number
+    },
+  ): Promise<Record<string, Provider.Model> | undefined> {
+    const timeout = options?.timeout ?? Number(Env.get("LITELLM_TIMEOUT") ?? "5000")
+    const url = `${host.replace(/\/+$/, "")}/model/info`
+
+    try {
+      const headers: Record<string, string> = {
+        "Content-Type": "application/json",
+        ...options?.headers,
+      }
+      if (options?.apiKey) {
+        headers["Authorization"] = `Bearer ${options.apiKey}`
+      }
+
+      const response = await fetch(url, {
+        headers,
+        signal: AbortSignal.timeout(timeout),
+      })
+
+      if (!response.ok) {
+        log.warn("LiteLLM model discovery failed", {
+          status: response.status,
+          url,
+        })
+        return undefined
+      }
+
+      const data = (await response.json()) as { data?: ModelInfoEntry[] }
+      const entries = data?.data
+      if (!Array.isArray(entries)) {
+        log.warn("LiteLLM /model/info returned unexpected format", { url })
+        return undefined
+      }
+
+      const models: Record<string, Provider.Model> = {}
+      for (const entry of entries) {
+        const model = toModel(entry)
+        if (model) {
+          models[model.id] = model
+        }
+      }
+
+      log.info("discovered models from LiteLLM proxy", {
+        count: Object.keys(models).length,
+        host,
+      })
+
+      return models
+    } catch (e) {
+      log.warn("LiteLLM model discovery error", { error: e, url })
+      return undefined
+    }
+  }
+}
diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index d94d0cbb223..09a328c35c8 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -41,6 +41,7 @@ import { createVercel } from "@ai-sdk/vercel"
 import { createGitLab, VERSION as GITLAB_PROVIDER_VERSION } from "@gitlab/gitlab-ai-provider"
 import { ProviderTransform } from "./transform"
 import { Installation } from "../installation"
+import { LiteLLM } from "./litellm"
 
 export namespace Provider {
   const log = Log.create({ service: "provider" })
@@ -589,6 +590,68 @@ export namespace Provider {
         },
       }
     },
+    litellm: async (provider) => {
+      const config = await Config.get()
+      const providerConfig = config.provider?.["litellm"]
+
+      const baseURL =
+        providerConfig?.options?.baseURL ??
+        Env.get("LITELLM_HOST") ??
+        Env.get("LITELLM_BASE_URL") ??
+        "http://localhost:4000"
+
+      const apiKey = await (async () => {
+        if (providerConfig?.options?.apiKey) return providerConfig.options.apiKey
+        const envKey = Env.get("LITELLM_API_KEY")
+        if (envKey) return envKey
+        const auth = await Auth.get("litellm")
+        if (auth?.type === "api") return auth.key
+        return undefined
+      })()
+
+      const customHeaders = iife(() => {
+        const raw = Env.get("LITELLM_CUSTOM_HEADERS")
+        if (!raw) return {}
+        try {
+          return JSON.parse(raw) as Record<string, string>
+        } catch {
+          return {}
+        }
+      })
+
+      const timeout = Number(Env.get("LITELLM_TIMEOUT") ?? "5000")
+
+      const discovered = await LiteLLM.discover(baseURL, {
+        apiKey,
+        headers: customHeaders,
+        timeout,
+      })
+
+      if (discovered) {
+        // Inject discovered models; user config models take precedence
+        for (const [modelID, model] of Object.entries(discovered)) {
+          if (!provider.models[modelID]) {
+            provider.models[modelID] = model
+          }
+        }
+      }
+
+      const hasModels = Object.keys(provider.models).length > 0
+      if (!hasModels) return { autoload: false }
+
+      return {
+        autoload: true,
+        options: {
+          baseURL,
+          apiKey,
+          litellmProxy: true,
+          ...customHeaders,
+        },
+        async getModel(sdk: any, modelID: string) {
+          return sdk.languageModel(modelID)
+        },
+      }
+    },
   }
 
   export const Model = z
@@ -795,6 +858,18 @@ export namespace Provider {
       }
     }
 
+    // Seed LiteLLM provider when env vars exist but no entry in database
+    if (!database["litellm"] && (Env.get("LITELLM_API_KEY") || Env.get("LITELLM_HOST") || Env.get("LITELLM_BASE_URL"))) {
+      database["litellm"] = {
+        id: "litellm",
+        name: "LiteLLM",
+        env: ["LITELLM_API_KEY"],
+        options: {},
+        source: "custom",
+        models: {},
+      }
+    }
+
     function mergeProvider(providerID: string, provider: Partial<Info>) {
       const existing = providers[providerID]
       if (existing) {
diff --git a/packages/opencode/src/provider/transform.ts b/packages/opencode/src/provider/transform.ts
index 759dab440d4..5749dcccbc1 100644
--- a/packages/opencode/src/provider/transform.ts
+++ b/packages/opencode/src/provider/transform.ts
@@ -359,6 +359,29 @@ export namespace ProviderTransform {
     }
     if (id.includes("grok")) return {}
 
+    // LiteLLM proxied models: infer reasoning variant from the underlying model
+    // to avoid false-positive reasoning param injection for aliased models
+    if (model.providerID === "litellm" && model.capabilities.reasoning) {
+      const apiId = model.api.id.toLowerCase()
+      if (apiId.includes("claude") || apiId.includes("anthropic")) {
+        return {
+          high: {
+            thinking: {
+              type: "enabled",
+              budgetTokens: Math.min(16_000, Math.floor(model.limit.output / 2 - 1)),
+            },
+          },
+          max: {
+            thinking: {
+              type: "enabled",
+              budgetTokens: Math.min(31_999, model.limit.output - 1),
+            },
+          },
+        }
+      }
+      return Object.fromEntries(WIDELY_SUPPORTED_EFFORTS.map((effort) => [effort, { reasoningEffort: effort }]))
+    }
+
     switch (model.api.npm) {
       case "@openrouter/ai-sdk-provider":
         if (!model.id.includes("gpt") && !model.id.includes("gemini-3") && !model.id.includes("claude")) return {}
diff --git a/packages/opencode/src/session/llm.ts b/packages/opencode/src/session/llm.ts
index fa880391276..5b95684ed94 100644
--- a/packages/opencode/src/session/llm.ts
+++ b/packages/opencode/src/session/llm.ts
@@ -160,6 +160,7 @@ export namespace LLM {
     // 1. Providers with "litellm" in their ID or API ID (auto-detected)
     // 2. Providers with explicit "litellmProxy: true" option (opt-in for custom gateways)
     const isLiteLLMProxy =
+      input.model.providerID === "litellm" ||
       provider.options?.["litellmProxy"] === true ||
       input.model.providerID.toLowerCase().includes("litellm") ||
       input.model.api.id.toLowerCase().includes("litellm")