chore: gemini-3.8-flash everywhere + auto-correct on Opus 5.5 (cost) (#1054)

## Why
GCP credits expired 2026-09-19; Vertex Gemini 3.1 Pro spend had grown $9
→ $37 → $105/30d. `gemini-3.8-flash` is $0.75/$3.75 per 1M (vs $2/$12
for 3.1 Pro; rises to $1.50/$7.50 on 2027-01-01). Opus 5.5 is $4/$20 (vs
$5/$25 for 4.8).

## Changes
- All Gemini defaults → `gemini-3.8-flash`: credit eval, custom lyrics,
auto-correct (default + compare leg), match judge, parse titles, tenant
bulk analyze, agentic AI router, error-monitor LLM analysis (also
`us-central1` → `global`, required for Gemini 3.x), frontend
`translate.py`.
- Auto-correct Claude leg `claude-opus-4-8` → `claude-opus-5-5` (ci.yml
env), with `effort: "high"` pinned (5.5 defaults to medium). Pricing
table gains both new models.
- Docs updated; version 0.241.1.

## Evidence (12 recent prod jobs, pre-review lyrics scored against the
human-reviewed final)
| Model | Precision | Word-level recall | $/job |
|---|---|---|---|
| gemini-3.1-pro-preview | 91% | 74% | $0.067 |
| gemini-3.8-flash | 98% | 38–41% | $0.020 |
| gemini-3.8-flash @ high thinking | 97% | 60% (4/12 timed out, 504) |
$0.041 |
| claude-opus-4-8 | 94% | 33% | $0.080 |
| claude-opus-5-5 (effort high) | 93% | 58% | $0.048 |

Flash for the Gemini leg is a deliberate cost/recall trade-off (Andrew
chose it); Opus 5.5 more than compensates. Also verified:
`gemini-3.8-flash` works on Vertex `global` (rejects
`thinking_level=minimal`; nothing here uses it), match-judge latency
median 3.3s / max 6.2s (< 12s timeout), Japanese translation smoke test
preserves ICU placeholders. Translation cache is keyed on English text +
locale, so no re-translation churn.

Tests: auto_correct, credit eval, custom lyrics, parse_titles,
error_monitor, agentic router suites pass locally.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **New Features**
* AI-powered correction, translation, matching, title parsing, and
evaluation now use updated models, including Gemini 3.8 Flash and Claude
Opus 5.5.
* **Updates**
* AI service pricing reflects scheduled rate changes, including a Gemini
3.8 Flash rate change beginning January 1, 2027.
  * Application version updated to 0.241.1.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Andrew Beveridge ☄️andClaude Opus 5.5 authored and GitHub committed 2026-09-26 14:53:04 -04:00
1 parent b177c36641
commit e9465c73d4
18 files changed
+72 -35

No files matched your search

+1 -1
View File
@@ -2025,7 +2025,7 @@ jobs:
--min-instances 2 \
--vpc-connector projects/nomadkaraoke/locations/us-central1/connectors/cloud-run-connector \
--vpc-egress private-ranges-only \
--set-env-vars "GOOGLE_CLOUD_PROJECT=${{ env.PROJECT_ID }},GCS_BUCKET_NAME=karaoke-gen-storage-${{ env.PROJECT_ID }},FIRESTORE_COLLECTION=jobs,ENVIRONMENT=production,KARAOKE_GEN_VERSION=$VERSION,DEFAULT_DROPBOX_PATH=/MediaUnsynced/Karaoke/Tracks-Organized,DEFAULT_GDRIVE_FOLDER_ID=1laRKAyxo0v817SstfM5XkpbWiNKNAMSX,ENABLE_CLOUD_TASKS=true,CLOUD_RUN_SERVICE_URL=https://api.nomadkaraoke.com,GCP_REGION=us-central1,DEFAULT_ENABLE_YOUTUBE_UPLOAD=true,DEFAULT_BRAND_PREFIX=NOMAD,DEFAULT_PRIVATE_DROPBOX_PATH=/MediaUnsynced/Karaoke/Tracks-NonPublished,DEFAULT_PRIVATE_BRAND_PREFIX=NOMADNP,USE_GCE_ENCODING=true,USE_GCE_PREVIEW_ENCODING=true,ENABLE_PUSH_NOTIFICATIONS=true,USE_CLOUD_RUN_JOBS_FOR_VIDEO=true,USE_CLOUD_RUN_JOBS_FOR_RENDER=true,EMAIL_PROVIDER=postmark,AUTO_CORRECT_MODEL=claude-opus-4-8,AUTO_CORRECT_COMPARE_MODELS=claude-opus-4-8;gemini-3.1-pro-preview,AUTO_CORRECT_PROACTIVE_ENABLED=true,AUTO_APPROVAL_BACKING_KEEP_ENABLED=true,COMMUNITY_DAILY_PICK_ENABLED=true,EDGE_AUTH_MODE=enforce,COMMIT_SHA=${{ steps.build-meta.outputs.sha }},PR_NUMBER=${{ steps.build-meta.outputs.pr_number }},PR_TITLE=${{ steps.build-meta.outputs.pr_title }}" \
--set-env-vars "GOOGLE_CLOUD_PROJECT=${{ env.PROJECT_ID }},GCS_BUCKET_NAME=karaoke-gen-storage-${{ env.PROJECT_ID }},FIRESTORE_COLLECTION=jobs,ENVIRONMENT=production,KARAOKE_GEN_VERSION=$VERSION,DEFAULT_DROPBOX_PATH=/MediaUnsynced/Karaoke/Tracks-Organized,DEFAULT_GDRIVE_FOLDER_ID=1laRKAyxo0v817SstfM5XkpbWiNKNAMSX,ENABLE_CLOUD_TASKS=true,CLOUD_RUN_SERVICE_URL=https://api.nomadkaraoke.com,GCP_REGION=us-central1,DEFAULT_ENABLE_YOUTUBE_UPLOAD=true,DEFAULT_BRAND_PREFIX=NOMAD,DEFAULT_PRIVATE_DROPBOX_PATH=/MediaUnsynced/Karaoke/Tracks-NonPublished,DEFAULT_PRIVATE_BRAND_PREFIX=NOMADNP,USE_GCE_ENCODING=true,USE_GCE_PREVIEW_ENCODING=true,ENABLE_PUSH_NOTIFICATIONS=true,USE_CLOUD_RUN_JOBS_FOR_VIDEO=true,USE_CLOUD_RUN_JOBS_FOR_RENDER=true,EMAIL_PROVIDER=postmark,AUTO_CORRECT_MODEL=claude-opus-5-5,AUTO_CORRECT_COMPARE_MODELS=claude-opus-5-5;gemini-3.8-flash,AUTO_CORRECT_PROACTIVE_ENABLED=true,AUTO_APPROVAL_BACKING_KEEP_ENABLED=true,COMMUNITY_DAILY_PICK_ENABLED=true,EDGE_AUTH_MODE=enforce,COMMIT_SHA=${{ steps.build-meta.outputs.sha }},PR_NUMBER=${{ steps.build-meta.outputs.pr_number }},PR_TITLE=${{ steps.build-meta.outputs.pr_title }}" \
--set-secrets "ANTHROPIC_API_KEY=anthropic-api-key:latest,AUDIOSHAKE_API_TOKEN=audioshake-api-key:latest,GENIUS_API_TOKEN=genius-api-key:latest,SPOTIFY_COOKIE_SP_DC=spotify-cookie:latest,RAPIDAPI_KEY=rapidapi-key:latest,DEFAULT_DISCORD_WEBHOOK_URL=discord-releases-webhook:latest,ADMIN_TOKENS=admin-tokens:latest,RED_API_KEY=red-api-key:latest,RED_API_URL=red-api-url:latest,OPS_API_KEY=ops-api-key:latest,OPS_API_URL=ops-api-url:latest,FLACFETCH_API_URL=flacfetch-api-url:latest,FLACFETCH_API_KEY=flacfetch-api-key:latest,POSTMARK_SERVER_TOKEN=postmark-server-token:latest,STRIPE_SECRET_KEY=stripe-secret-key:latest,STRIPE_WEBHOOK_SECRET=stripe-webhook-secret:latest,LANGFUSE_PUBLIC_KEY=langfuse-public-key:latest,LANGFUSE_SECRET_KEY=langfuse-secret-key:latest,LANGFUSE_HOST=langfuse-host:latest,ENCODING_WORKER_URL=encoding-worker-url:latest,ENCODING_WORKER_API_KEY=encoding-worker-api-key:latest,ENCODING_WORKER_FALLBACK_VMS=encoding-worker-fallback-vms:latest,VAPID_PUBLIC_KEY=vapid-public-key:latest,VAPID_PRIVATE_KEY=vapid-private-key:latest,E2E_BYPASS_KEY=e2e-bypass-key:latest,GDRIVE_VALIDATOR_URL=gdrive-validator-url:latest,EDGE_ORIGIN_SECRET=edge-origin-secret:latest,KARAOKEHUNT_FORWARDER_SECRET=karaokehunt-forwarder-secret:latest,KJBOX_PARTNER_SECRET=kjbox-partner-secret:latest"
# R3/R4 — smoke the tagged candidate revision BEFORE it serves any traffic.
+1 -1
View File
@@ -112,7 +112,7 @@ See `docs/TESTING.md` § "Ad-Hoc Production Debugging" for full details.
- Only `/admin/*` is English-only — it has no `[locale]/admin` counterpart and inherits `DefaultIntlProvider` from the root layout
- Internal links use `Link` from `@/i18n/routing` (locale-aware)
- After adding/changing English strings: `python frontend/scripts/translate.py --messages-dir frontend/messages --target all`
- Translation uses Gemini 3.1 Pro via Vertex AI with GCS cache for efficiency
- Translation uses Gemini 3.8 Flash via Vertex AI with GCS cache for efficiency
- CI validates all locale files have matching keys — **PR will fail if translations are missing**
- Pre-commit hook auto-translates when en.json changes (enable: `git config core.hooksPath .githooks`)
- Don't hardcode user-facing strings — add to `messages/en.json` and use `t('key')`
+6 -6
View File
@@ -84,7 +84,7 @@ class Settings(BaseSettings):
# Only used when skip_correction=false
# When enabled, uses Gemini via Vertex AI for intelligent lyrics correction
use_agentic_ai: bool = os.getenv("USE_AGENTIC_AI", "true").lower() in ("true", "1", "yes")
agentic_ai_model: str = os.getenv("AGENTIC_AI_MODEL", "vertexai/gemini-3-flash-preview")
agentic_ai_model: str = os.getenv("AGENTIC_AI_MODEL", "vertexai/gemini-3.8-flash")
# Timeout for agentic correction in seconds. If correction takes longer, abort and
# use uncorrected transcription - human review will fix any issues.
agentic_correction_timeout_seconds: int = int(os.getenv("AGENTIC_CORRECTION_TIMEOUT_SECONDS", "180"))
@@ -105,12 +105,12 @@ class Settings(BaseSettings):
# before granting. Fail-CLOSED: if evaluation fails, the decision becomes
# "pending_review" (no auto-grant; an admin is notified to decide).
credit_eval_enabled: bool = os.getenv("CREDIT_EVAL_ENABLED", "true").lower() in ("true", "1", "yes")
credit_eval_model: str = os.getenv("CREDIT_EVAL_MODEL", "gemini-3.1-pro-preview")
credit_eval_model: str = os.getenv("CREDIT_EVAL_MODEL", "gemini-3.8-flash")
# Custom Lyrics (LLM-powered) — used by the Custom Lyrics mode in the
# Edit All Lyrics modal. Stateless service: takes operator-provided
# custom-lyrics text/file and returns transformed lyric lines.
custom_lyrics_model: str = os.getenv("CUSTOM_LYRICS_MODEL", "gemini-3.1-pro-preview")
custom_lyrics_model: str = os.getenv("CUSTOM_LYRICS_MODEL", "gemini-3.8-flash")
custom_lyrics_max_file_mb: int = int(os.getenv("CUSTOM_LYRICS_MAX_FILE_MB", "5"))
custom_lyrics_max_input_lines: int = int(os.getenv("CUSTOM_LYRICS_MAX_INPUT_LINES", "500"))
custom_lyrics_max_iterations: int = int(os.getenv("CUSTOM_LYRICS_MAX_ITERATIONS", "4"))
@@ -120,7 +120,7 @@ class Settings(BaseSettings):
# AI auto-correct suggestions — opt-in, user-triggered from the lyrics
# review UI. Stateless: one whole-song LLM call returning word-level
# suggestions the reviewer accepts/rejects individually.
auto_correct_model: str = os.getenv("AUTO_CORRECT_MODEL", "gemini-3.1-pro-preview")
auto_correct_model: str = os.getenv("AUTO_CORRECT_MODEL", "gemini-3.8-flash")
# Models used by the multi-model "compare" mode (semicolon-separated —
# Cloud Run --set-env-vars is comma-delimited). Empty = single-model only.
auto_correct_compare_models: str = os.getenv("AUTO_CORRECT_COMPARE_MODELS", "")
@@ -175,7 +175,7 @@ class Settings(BaseSettings):
match_judge_enabled: bool = os.getenv("MATCH_JUDGE_ENABLED", "true").lower() in (
"true", "1", "yes",
)
match_judge_model: str = os.getenv("MATCH_JUDGE_MODEL", "gemini-3.5-flash")
match_judge_model: str = os.getenv("MATCH_JUDGE_MODEL", "gemini-3.8-flash")
# Bounds the AI call so a hung model never strands the submission UI. Kept
# generous because Vertex returns 504 DEADLINE_EXCEEDED if generation can't
# finish within this deadline (3s was too tight for gemini-3.5-flash); the
@@ -187,7 +187,7 @@ class Settings(BaseSettings):
parse_titles_enabled: bool = os.getenv("PARSE_TITLES_ENABLED", "true").lower() in (
"1", "true", "yes",
)
parse_titles_model: str = os.getenv("PARSE_TITLES_MODEL", "gemini-3.5-flash")
parse_titles_model: str = os.getenv("PARSE_TITLES_MODEL", "gemini-3.8-flash")
parse_titles_timeout_ms: int = int(os.getenv("PARSE_TITLES_TIMEOUT_MS", "20000"))
parse_titles_max_items: int = int(os.getenv("PARSE_TITLES_MAX_ITEMS", "200"))
# Items per Gemini call: large client batches are split into concurrent
+25 -4
View File
@@ -22,6 +22,7 @@ logged tokens rather than trusting the historical USD value.
"""
from __future__ import annotations
from datetime import date, datetime, timezone
from typing import Optional, TypedDict
@@ -49,27 +50,47 @@ class _Rate(TypedDict):
MODEL_PRICING: dict[str, _Rate] = {
# --- Anthropic (authoritative) ---
"claude-fable-5": {"input": 10.0, "output": 50.0, "cache_read": 1.0},
"claude-opus-5-5": {"input": 4.0, "output": 20.0, "cache_read": 0.2},
"claude-opus-4-8": {"input": 5.0, "output": 25.0, "cache_read": 0.5},
"claude-sonnet-4-6": {"input": 3.0, "output": 15.0, "cache_read": 0.3},
"claude-haiku-4-5": {"input": 1.0, "output": 5.0, "cache_read": 0.1},
# --- Gemini (ESTIMATE — Vertex 3.x Pro preview, <=200K-token context
# tier; confirm against current Google pricing before trusting USD) ---
"gemini-3.1-pro-preview": {"input": 2.0, "output": 12.0, "cache_read": 0.2},
# Promo rate through 2026-12-31; see _RATE_CHANGES for the standard rate.
"gemini-3.8-flash": {"input": 0.75, "output": 3.75, "cache_read": 0.075},
}
# Scheduled price changes: model -> [(effective_date, rate)], oldest first.
# The latest entry whose date has arrived overrides MODEL_PRICING.
_RATE_CHANGES: dict[str, list[tuple[date, _Rate]]] = {
"gemini-3.8-flash": [
(date(2027, 1, 1), {"input": 1.5, "output": 7.5, "cache_read": 0.15}),
],
}
def get_rate(model: str) -> Optional[_Rate]:
def _effective(model: str, rate: _Rate, on: Optional[date]) -> _Rate:
on = on or datetime.now(timezone.utc).date()
for effective, changed in _RATE_CHANGES.get(model, []):
if on >= effective:
rate = changed
return rate
def get_rate(model: str, on: Optional[date] = None) -> Optional[_Rate]:
"""Return the rate card for ``model``, or None if we have no pricing.
Falls back to a prefix match on the model family (e.g. a dated
``claude-fable-5-2026...`` snapshot still resolves to ``claude-fable-5``)
so a minor model-id change doesn't silently drop cost tracking.
so a minor model-id change doesn't silently drop cost tracking. ``on``
(default: today, UTC) selects any scheduled price change.
"""
if model in MODEL_PRICING:
return MODEL_PRICING[model]
return _effective(model, MODEL_PRICING[model], on)
for known, rate in MODEL_PRICING.items():
if model.startswith(known):
return rate
return _effective(known, rate, on)
return None
+5 -1
View File
@@ -674,7 +674,11 @@ class AutoCorrectService:
system=system_prompt,
messages=[{"role": "user", "content": user_prompt}],
thinking={"type": "adaptive"},
output_config={"format": {"type": "json_schema", "schema": strict_schema}},
# Pin effort: Opus 5.5 defaults to "medium" (4.8 defaulted to "high").
output_config={
"effort": "high",
"format": {"type": "json_schema", "schema": strict_schema},
},
)
except Exception as exc:
if _is_transient_model_error(exc):
+2 -2
View File
@@ -98,8 +98,8 @@ Shared infrastructure:
# LLM analysis settings
# ---------------------------------------------------------------------------
LLM_ANALYSIS_MODEL: str = os.environ.get("LLM_ANALYSIS_MODEL", "gemini-2.0-flash-001")
LLM_VERTEX_LOCATION: str = os.environ.get("LLM_VERTEX_LOCATION", "us-central1")
LLM_ANALYSIS_MODEL: str = os.environ.get("LLM_ANALYSIS_MODEL", "gemini-3.8-flash")
LLM_VERTEX_LOCATION: str = os.environ.get("LLM_VERTEX_LOCATION", "global")
#: Minimum number of active error patterns required before triggering LLM analysis.
MIN_PATTERNS_FOR_ANALYSIS: int = 3
+2 -2
View File
@@ -70,9 +70,9 @@ RESPONSE_SCHEMA = {
def _model() -> str:
try:
from backend.config import settings
return getattr(settings, "match_judge_model", "gemini-3.5-flash")
return getattr(settings, "match_judge_model", "gemini-3.8-flash")
except Exception: # pragma: no cover - config import guard
return "gemini-3.5-flash"
return "gemini-3.8-flash"
def build_prompts(
+2 -2
View File
@@ -59,9 +59,9 @@ RESPONSE_SCHEMA = {
def _model() -> str:
try:
from backend.config import settings
return getattr(settings, "parse_titles_model", "gemini-3.5-flash")
return getattr(settings, "parse_titles_model", "gemini-3.8-flash")
except Exception: # pragma: no cover - config guard
return "gemini-3.5-flash"
return "gemini-3.8-flash"
def build_prompts(items: list[dict]) -> tuple[str, str]:
+2 -2
View File
@@ -445,9 +445,9 @@ def _default_model() -> str:
try:
from backend.config import settings
return getattr(settings, "match_judge_model", "gemini-3.5-flash")
return getattr(settings, "match_judge_model", "gemini-3.8-flash")
except Exception: # pragma: no cover - config import guard
return "gemini-3.5-flash"
return "gemini-3.8-flash"
def default_generate(system_prompt: str, user_prompt: str) -> dict:
@@ -171,3 +171,15 @@ def test_suggest_logs_unavailable_usage_without_crashing(caplog) -> None:
)
assert "auto-correct usage job=job-2 model=claude-fable-5 tokens=unavailable" in caplog.text
assert "total_est_cost_usd=unknown" in caplog.text
def test_gemini_38_flash_rate_switches_to_standard_on_2027_01_01() -> None:
from datetime import date
from backend.services.auto_correct.pricing import get_rate
assert get_rate("gemini-3.8-flash", on=date(2026, 12, 31))["output"] == 3.75
assert get_rate("gemini-3.8-flash", on=date(2027, 1, 1))["output"] == 7.5
assert get_rate("gemini-3.8-flash", on=date(2027, 1, 1))["input"] == 1.5
# Models without scheduled changes are unaffected.
assert get_rate("claude-opus-5-5", on=date(2027, 6, 1))["output"] == 20.0
+1 -1
View File
@@ -8,7 +8,7 @@ def test_settings_parse_titles_defaults():
from backend.config import Settings
s = Settings()
assert s.parse_titles_enabled is True
assert s.parse_titles_model == "gemini-3.5-flash"
assert s.parse_titles_model == "gemini-3.8-flash"
assert s.parse_titles_timeout_ms == 20000
assert s.parse_titles_max_items == 200
+2 -2
View File
@@ -77,7 +77,7 @@ The verify page (`/auth/verify`) shows these states in sequence:
| Environment Variable | Default | Description |
|---------------------|---------|-------------|
| `CREDIT_EVAL_ENABLED` | `true` | Set to `false` to disable AI evaluation and auto-grant all credits |
| `CREDIT_EVAL_MODEL` | `gemini-3.1-pro-preview` | Gemini model to use for evaluation |
| `CREDIT_EVAL_MODEL` | `gemini-3.8-flash` | Gemini model to use for evaluation |
### Emergency Rollback
@@ -96,7 +96,7 @@ Every evaluation is logged to the `credit_evaluations` Firestore collection:
"reasoning": "No suspicious correlations found — clean user",
"confidence": 1.0,
"error": null,
"model": "gemini-3.1-pro-preview",
"model": "gemini-3.8-flash",
"signals_snapshot": {
"fingerprint_match_count": 0,
"ip_match_count": 0,
+3 -3
View File
@@ -689,13 +689,13 @@ Returns:
"reason": "All references read 'chlorine' here",
"category": "mishearing",
"confidence": 0.95,
"models": ["claude-opus-4-8", "gemini-3.1-pro-preview"],
"models": ["claude-opus-5-5", "gemini-3.8-flash"],
"consensus": 2,
"total_models": 2,
"conflict_group": null
}
],
"model": "claude-opus-4-8, gemini-3.1-pro-preview",
"model": "claude-opus-5-5, gemini-3.8-flash",
"elapsed_seconds": 27.4,
"settings_applied": { "suggest_adlib_removal": true, "allow_insertions": true, "min_confidence": 0.0 },
"warnings": [],
@@ -932,7 +932,7 @@ Content-Type: application/json
Decides the official formatting for a typed artist/title and whether it matches a
real song. A deterministic normalizer + the catalog run first; a light Vertex
Gemini model (`MATCH_JUDGE_MODEL`, default `gemini-3.5-flash`) is consulted
Gemini model (`MATCH_JUDGE_MODEL`, default `gemini-3.8-flash`) is consulted
**only** when those aren't confident. `audio_confidence_tier` (1=strong..3=weak,
optional) lets the judge tell whether weak audio results hint at a typo. The call
never blocks job creation — on timeout/error it returns a `none` verdict. Disable
+1 -1
View File
@@ -251,7 +251,7 @@ LyricsTranscriber LyricsTranscriber
|---------|---------|----------|
| Cloud Run GPU Job (L4, us-east4) | Audio stem separation — `Separator` runs directly in the audio worker Job (no HTTP hop) | Yes |
| AudioShake | Lyrics transcription | Yes |
| Vertex AI | Agentic AI correction (Gemini 3 Flash) | Default off (SKIP_CORRECTION=false to enable) |
| Vertex AI | Agentic AI correction (Gemini 3.8 Flash) | Default off (SKIP_CORRECTION=false to enable) |
| Genius | Reference lyrics | Yes |
| Flacfetch | Audio downloads (YouTube, torrents) | Recommended* |
| YouTube API | Video upload | Optional |
+1 -1
View File
@@ -39,7 +39,7 @@ from translation_cache import TranslationCache
from google import genai
from google.genai import types
MODEL = "gemini-3.1-pro-preview"
MODEL = "gemini-3.8-flash"
PROJECT = "nomadkaraoke"
LOCATION = "global"
@@ -7,7 +7,7 @@ from .providers.config import ProviderConfig
# Default model for cloud deployments - Gemini 3 Flash via Vertex AI
# Note: Gemini 3 models require 'global' location (not regional like us-central1)
DEFAULT_CLOUD_MODEL = "vertexai/gemini-3-flash-preview"
DEFAULT_CLOUD_MODEL = "vertexai/gemini-3.8-flash"
class ModelRouter:
@@ -20,7 +20,7 @@ class ModelRouter:
"""Choose appropriate model based on gap characteristics.
Returns model identifier in format "provider/model" for LangChain:
- "vertexai/gemini-3-flash-preview" for Gemini via Vertex AI (default)
- "vertexai/gemini-3.8-flash" for Gemini via Vertex AI (default)
- "ollama/llama3.2:latest" for local Ollama models
- "openai/gpt-4" for OpenAI models
- "anthropic/claude-3-sonnet-20240229" for Anthropic models
@@ -34,7 +34,7 @@ class ModelRouter:
if self._config.privacy_mode:
return "ollama/llama3.2:latest"
# Default to Gemini 3 Flash for all cases (fast, cost-effective, latest capabilities)
# Default to Gemini 3.8 Flash for all cases (fast, cost-effective, latest capabilities)
return DEFAULT_CLOUD_MODEL
+1 -1
View File
@@ -1,6 +1,6 @@
[tool.poetry]
name = "karaoke-gen"
version = "0.241.0"
version = "0.241.1"
description = "Generate karaoke videos with synchronized lyrics. Handles the entire process from downloading audio and lyrics to creating the final video with title screens."
authors = ["Andrew Beveridge <andrew@beveridge.uk>"]
license = "MIT"
+2 -2
View File
@@ -19,7 +19,7 @@ Usage:
python scripts/measure_autocorrect_cost.py --jobs abc123 def456
# Single-model instead of the multi-model compare config
python scripts/measure_autocorrect_cost.py --limit 3 --models claude-opus-4-8
python scripts/measure_autocorrect_cost.py --limit 3 --models claude-opus-5-5
Requires:
- ANTHROPIC_API_KEY in the environment (for claude-* models)
@@ -55,7 +55,7 @@ logger.setLevel(logging.INFO)
PROJECT = "nomadkaraoke"
GCS_BUCKET = "karaoke-gen-storage-nomadkaraoke"
DEFAULT_MODELS = ["claude-opus-4-8", "gemini-3.1-pro-preview"]
DEFAULT_MODELS = ["claude-opus-5-5", "gemini-3.8-flash"]
# ---------------------------------------------------------------------------