Update default models

.github/workflows/ci.yml

@@ -36,8 +36,10 @@ jobs:
           issue-numbers: "3 6"
           prompt-path: examples/AutoTriage.prompt
           enabled: false
-          model-fast: ''
-          model-pro: gemini-flash-latest # use flash for pro model and skip fast pass due to rate limits
+          model-fast: gemini-2.5-flash
+          model-fast-temperature: 0.0
+          model-pro: gemini-3-flash-preview
+          model-pro-temperature: 1.0
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}

README.md

@@ -5,7 +5,8 @@ Keep issues and pull requests moving: reads the latest context, drafts the next
 ## How it works
 
 - The run starts with a fast AI pass to gather signals, summarize the thread, and draft the intended operations.
-- A reviewing AI pass (default: `gemini-2.5-pro`) replays the plan and confirms labels, comments, etc, before anything is written.
+- A reviewing AI pass (default: `gemini-3-flash-preview`) replays the plan and confirms labels, comments, etc, before anything is written.
+- Defaults use the free-tier models (`gemini-2.5-flash` + `gemini-3-flash-preview`) rather than `gemini-3-pro`.
 - The full thought process along with all actions can be inspected in the workflow artifacts.
 - It will keep going until it runs out of issues or tokens, or reaches the specified limit.
 
@@ -43,10 +44,10 @@ jobs:
 | `readme-path` | Extra Markdown context uploaded to the AI prompt. | `README.md` |
 | `enabled` | `"true"` applies changes, `"false"` logs the plan only. | `"true"` |
 | `db-path` | Persist per-item history between runs. | - |
-| `model-fast` | Fast analysis model for the first pass. Leave blank to skip. | bundled fast model |
-| `model-pro` | Review model that double-checks uncertain plans. | bundled review model |
+| `model-fast` | Fast analysis model for the first pass. Leave blank to skip. | `gemini-2.5-flash` |
+| `model-pro` | Review model that double-checks uncertain plans. | `gemini-3-flash-preview` |
 | `model-fast-temperature` | Sampling temperature for fast model (`0` deterministic -> `2` exploratory). | `0.0` |
-| `model-pro-temperature` | Sampling temperature for pro model (`0` deterministic -> `2` exploratory). Gemini 3.0 recommends `1.0`. | `1.0` |
+| `model-pro-temperature` | Sampling temperature for pro model (`0` deterministic -> `2` exploratory). Gemini 3 recommends `1.0`. | `1.0` |
 | `max-timeline-events` | Maximum recent timeline events included in the prompt. | `40` |
 | `max-triages` | Cap on items that escalate to the review pass per run. | `20` |
 | `max-fast-runs` | Cap on items analyzed with the fast model per run. | `100` |

action.yml

@@ -30,19 +30,19 @@ inputs:
   model-fast:
     description: "Identifier of the fast analysis model used for the first pass (default shown below). Leave blank to skip the fast pass."
     required: false
-    default: "gemini-flash-latest"
+    default: "gemini-2.5-flash"
   model-pro:
     description: "Identifier of the deeper review model used when the fast pass asks for confirmation."
     required: false
-    default: "gemini-2.5-pro"
+    default: "gemini-3-flash-preview"
   model-fast-temperature:
     description: "Sampling temperature for the fast model (0-2); lower keeps responses deterministic while higher explores alternatives."
     required: false
     default: "0.0"
   model-pro-temperature:
-    description: "Sampling temperature for the pro model (0-2); Gemini 3.0 recommends 1.0 for optimal performance."
+    description: "Sampling temperature for the pro model (0-2); Gemini 3 recommends 1.0 for optimal performance."
     required: false
-    default: "0.0"
+    default: "1.0"
   max-timeline-events:
     description: "Maximum number of recent timeline events injected into the prompt context."
     required: false

src/env.ts

@@ -42,17 +42,17 @@ export function getConfig(): Config {
   const readmePath = core.getInput('readme-path') || 'README.md';
   const dbPath = core.getInput('db-path');
   const modelFastInput = core.getInput('model-fast');
-  const modelFast = modelFastInput || 'gemini-flash-latest';
+  const modelFast = modelFastInput || 'gemini-2.5-flash';
   const skipFastPass = modelFastInput === '';
-  const modelPro = core.getInput('model-pro') || 'gemini-2.5-pro';
+  const modelPro = core.getInput('model-pro') || 'gemini-3-flash-preview';
   const fastTemperatureInput = core.getInput('model-fast-temperature');
   const parsedFastTemperature = Number(
-    fastTemperatureInput === undefined || fastTemperatureInput === '' ? '0' : fastTemperatureInput
+    fastTemperatureInput === undefined || fastTemperatureInput === '' ? '0.0' : fastTemperatureInput
   );
   const modelFastTemperature = Number.isFinite(parsedFastTemperature) ? parsedFastTemperature : 0;
   const proTemperatureInput = core.getInput('model-pro-temperature');
   const parsedProTemperature = Number(
-    proTemperatureInput === undefined || proTemperatureInput === '' ? '0' : proTemperatureInput
+    proTemperatureInput === undefined || proTemperatureInput === '' ? '1.0' : proTemperatureInput
   );
   const modelProTemperature = Number.isFinite(parsedProTemperature) ? parsedProTemperature : 0;
   const thinkingBudget = -1;

src/gemini.ts

@@ -51,7 +51,7 @@ export class GeminiClient {
 
   private async parseJson<T>(response: GenerateContentResponse): Promise<{ data: T; thoughts: string; inputTokens: number; outputTokens: number }> {
     // Manually extract text from parts to avoid warnings about non-text parts (e.g., thoughtSignature)
-    // when using Gemini 3.0 models with thinking enabled
+    // when using Gemini 3 models with thinking enabled
     const thoughts: string[] = [];
     const textParts: string[] = [];
 

tests/geminiSdk.test.ts

@@ -3,7 +3,7 @@ import { GeminiClient, buildJsonPayload } from '../src/gemini';
 
 describe('Gemini (real API)', () => {
     const apiKey = process.env.GEMINI_API_KEY;
-    const model = 'gemini-flash-latest';
+    const model = 'gemini-2.5-flash';
 
     if (!apiKey) {
         console.warn('GEMINI_API_KEY environment variable must be set to run Gemini tests. Skipping Gemini tests.');