feat(plugin): surface reasoning-effort variants from /v1/model/info reported by LiteLLM

CHANGELOG.md

@@ -26,6 +26,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Features
+
+* **plugin:** surface reasoning-effort variants from `/v1/model/info` reported by LiteLLM
+
 ### Fixed
 - **Embedding / image / audio models no longer appear in the OpenCode
   model picker.** The non-chat filter in `toConfigModel()` was a dead

README.md

@@ -193,6 +193,13 @@ detects reasoning-tier models from their id (`gpt-5*`, `o1*`, `o3*`,
 `o4*`) and from LiteLLM's `mode === 'responses'` field, and creates the
 sibling provider lazily.
 
+If LiteLLM reports per-model reasoning-effort support (e.g.
+`supports_low_reasoning_effort`, `supports_medium_reasoning_effort`,
+`supports_high_reasoning_effort` in `model_info`), the plugin automatically
+surfaces those as OpenCode variants under the discovered model. Each variant
+sets `reasoningEffort` to the reported level, so you can switch between
+effort levels from the model picker without hand-curating every entry.
+
 To override the routing per model:
 
 ```jsonc

src/plugin/build-model.ts

@@ -1,5 +1,5 @@
 import type { Model as ModelV2 } from '@opencode-ai/sdk/v2'
-import type { LiteLLMModel } from '../types'
+import type { LiteLLMModel, LiteLLMModelInfo } from '../types'
 import {
   categorizeModel,
   formatModelName,
@@ -20,6 +20,7 @@ export function buildModelV2(
   providerID: string,
   api: { id: string; url: string; npm: string },
   model: LiteLLMModel,
+  info?: LiteLLMModelInfo,
 ): ModelV2 {
   const type = categorizeModel(model)
   const isVision = !!model.supports_vision
@@ -66,5 +67,13 @@ export function buildModelV2(
     options: {},
     headers: {},
     release_date: '',
+    variants: info?.supports_reasoning_efforts?.length
+      ? Object.fromEntries(
+          info.supports_reasoning_efforts.map((effort) => [
+            effort,
+            { reasoningEffort: effort },
+          ]),
+        )
+      : undefined,
   }
 }

src/plugin/discover.ts

@@ -2,11 +2,12 @@ import type { Model as ModelV2, Provider as ProviderV2 } from '@opencode-ai/sdk/
 import {
   autoDetectLiteLLM,
   checkLiteLLMHealth,
+  discoverLiteLLMModelInfo,
   discoverLiteLLMModels,
   normalizeBaseURL,
 } from '../utils/litellm-api'
 import { requiresResponsesAPI } from '../utils/format-model-name'
-import type { LiteLLMModel, Transport, TransportPolicy } from '../types'
+import type { LiteLLMModel, LiteLLMModelInfo, Transport, TransportPolicy } from '../types'
 import { buildModelV2 } from './build-model'
 
 const DISCOVERY_TIMEOUT_MS = 5000
@@ -138,16 +139,29 @@ export async function discoverBucket(
       return
     }
 
-    let models: LiteLLMModel[]
-    try {
-      models = await discoverLiteLLMModels(baseURL, apiKey, customHeaders)
-    } catch (error) {
+    const [modelsResult, infoResult] = await Promise.allSettled([
+      discoverLiteLLMModels(baseURL, apiKey, customHeaders),
+      discoverLiteLLMModelInfo(baseURL, apiKey, customHeaders),
+    ])
+
+    if (modelsResult.status === 'rejected') {
       console.warn(
         '[opencode-litellm] Model discovery failed:',
-        error instanceof Error ? error.message : String(error),
+        modelsResult.reason instanceof Error ? modelsResult.reason.message : String(modelsResult.reason),
       )
       return
     }
+    const models = modelsResult.value
+
+    let infoByName: Map<string, LiteLLMModelInfo> | null = null
+    if (infoResult.status === 'fulfilled') {
+      infoByName = infoResult.value
+    } else {
+      console.warn(
+        '[opencode-litellm] /v1/model/info unavailable; model costs will be zero:',
+        infoResult.reason instanceof Error ? infoResult.reason.message : String(infoResult.reason),
+      )
+    }
 
     if (models.length === 0) {
       console.warn(
@@ -176,7 +190,8 @@ export async function discoverBucket(
       // it with "team not allowed"). Set `api.id` per-model so each
       // entry carries the correct upstream model name.
       const perModelApi = { ...resolvedApi, id: model.id }
-      out[model.id] = buildModelV2(resolvedApi.id, perModelApi, model)
+      const info = infoByName?.get(model.id) ?? {}
+      out[model.id] = buildModelV2(resolvedApi.id, perModelApi, model, info)
     }
   }
 

src/plugin/index.ts

@@ -87,7 +87,10 @@ function enrichModel(model: LiteLLMModel, info: LiteLLMModelInfo): LiteLLMModel
  * image, audio) — they can't be used as primary chat models and would
  * clutter the picker.
  */
-function toConfigModel(model: LiteLLMModel): Record<string, unknown> | null {
+function toConfigModel(
+  model: LiteLLMModel,
+  info?: LiteLLMModelInfo,
+): Record<string, unknown> | null {
   const type = categorizeModel(model)
   if (type === 'embedding' || type === 'image' || type === 'audio') {
     return null
@@ -117,6 +120,14 @@ function toConfigModel(model: LiteLLMModel): Record<string, unknown> | null {
   if (input.length > 1) {
     entry.modalities = { input, output: ['text'] }
   }
+  entry.variants = info?.supports_reasoning_efforts?.length
+    ? Object.fromEntries(
+        info.supports_reasoning_efforts.map((effort) => [
+          effort,
+          { reasoningEffort: effort },
+        ]),
+      )
+    : undefined
   return entry
 }
 
@@ -303,7 +314,10 @@ export const LiteLLMPlugin: Plugin = async (_input: PluginInput) => {
             if (models[model.id]) continue
             const info = infoByName?.get(model.id)
             if (infoByName && !info) unmatched.push(model.id)
-            const entry = toConfigModel(info ? enrichModel(model, info) : model)
+            const entry = toConfigModel(
+              info ? enrichModel(model, info) : model,
+              info,
+            )
             if (!entry) {
               skipped++
               continue

src/types/index.ts

@@ -56,6 +56,7 @@ export interface LiteLLMModelInfo {
   supports_function_calling?: boolean
   supports_vision?: boolean
   supports_reasoning?: boolean
+  supports_reasoning_efforts?: string[]
   supports_pdf_input?: boolean
   supports_audio_input?: boolean
 }

src/utils/litellm-api.ts

@@ -125,6 +125,25 @@ export async function discoverLiteLLMModelInfo(
         info[flag] = paramsValue
       }
     }
+
+    // Add supports reasoning efforts if present in litellm_params
+    const reasoningEffortPattern = /^supports_([a-z]+)_reasoning_effort$/
+    const efforts = new Set<string>()
+    for (const source of [info, entry.litellm_params]) {
+      if (!source || typeof source !== 'object') continue
+      for (const [key, value] of Object.entries(source)) {
+        const match = key.match(reasoningEffortPattern)
+        if (match && value === true) {
+          efforts.add(match[1])
+        }
+      }
+    }
+    if (efforts.size > 0) {
+      info.supports_reasoning_efforts = [
+        ...new Set([...(info.supports_reasoning_efforts ?? []), ...efforts]),
+      ]
+    }
+
     // Index under every alias LiteLLM may use for this model — the
     // `/v1/models` id can match any of them depending on how the
     // deployment names its entries (alias vs upstream model string).