fix(google): cap max_tokens at 65536 on the Gemini 3.8 Live models

CHANGELOG.md

@@ -23,6 +23,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Images** — OpenAI `gpt-image-2`, `gpt-image-2.5-flare` and `gpt-image-2.5-sunburst` accept any documented `WIDTHxHEIGHT` size (multiples of 16, 1:3 to 3:1, at most 3840 per edge, 655,360–8,294,400 pixels) with the standard sizes kept as presets, instead of eight fixed sizes (#564).
 - **Text** — Anthropic on Claude on Google Cloud (GoogleADC) sends image and document inputs as base64 sources; the route accepts no URL sources (#570).
 - **Videos** — Google `veo-3.1-generate-preview` is no longer registered: the Gemini API shut it down on October 22, 2026 and now answers 404 `NOT_FOUND`. No Veo model remains registered; Google's replacement is `gemini-omni-1.2-flash` (#585).
+- **Live** — Google `gemini-3.8-live` and `gemini-3.8-live-extended-thinking` constrain `max_tokens` to 1–65,536, the documented output-token limit, as the Google text models do; a larger value raises `ConstraintViolationError` instead of being sent (#590).
 
 ---
 

src/celeste/modalities/live/providers/google/models.py

@@ -5,6 +5,7 @@
     Choice,
     Constraint,
     ImagesConstraint,
+    Range,
     ToolSupport,
 )
 from celeste.core import Modality, Operation, Provider
@@ -33,6 +34,7 @@
         parameter_constraints={
             **_MEDIA_CONSTRAINTS,
             LiveParameter.VOICE: VoiceConstraint(voices=GOOGLE_VOICES),
+            LiveParameter.MAX_TOKENS: Range(min=1, max=65536),
         },
     ),
     Model(
@@ -44,6 +46,7 @@
         parameter_constraints={
             **_MEDIA_CONSTRAINTS,
             LiveParameter.VOICE: VoiceConstraint(voices=GOOGLE_VOICES),
+            LiveParameter.MAX_TOKENS: Range(min=1, max=65536),
             LiveParameter.THINKING_LEVEL: Choice(options=["LOW", "MEDIUM", "HIGH"]),
         },
     ),