luke/ai-podcast
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update model routing with latest OpenRouter models

Browse Source 
Style-matched defaults:
- Grok 4.1 Fast for edgy callers (high_energy, confrontational, comedian etc.)
- Claude Sonnet 4.6 for emotional callers (quiet_nervous, sweet_earnest, emotional)
- Mistral Large 2512 for deadpan/mysterious/world-weary
- DeepSeek R1 Distill for storyteller/oversharer/conspiracy/rambler
- Gemini 2.5 Flash for know_it_all
- Llama 3.3 70B for first_time/reluctant callers

Category routing: Grok 4.1 Fast for dialog/devon/backgrounds, Gemini Flash for monitor/summary
Updated OPENROUTER_MODELS and OPENROUTER_PRICING with all new models

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
tcpsyn
2026-03-21 02:31:33 -06:00
parent e614599650
commit c516402402
4 changed files with 80 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/config.py
+10 -11
View File
@@ -29,21 +29,20 @@ class Settings(BaseSettings):
# LLM Settings
llm_provider: str = "openrouter" # "openrouter" or "ollama"
openrouter_model: str = "anthropic/claude-sonnet-4-5" # primary/default model
openrouter_model: str = "anthropic/claude-sonnet-4.6" # primary/default model
ollama_model: str = "llama3.2"
ollama_host: str = "http://localhost:11434"
# Per-category model routing — cheaper models for non-critical tasks
# Categories: caller_dialog, devon_monitor, devon_ask, background_gen,
# call_summary, news_summary, topic_gen, unknown
# Per-category model routing
# caller_dialog is overridden by style_matched routing (see Session.caller_model_map)
category_models: dict = {
"caller_dialog": "x-ai/grok-4", # full Grok 4 — edgier dialog, latency OK (gaps cut in post)
"devon_ask": "x-ai/grok-4", # Devon should match the show's edgy energy
"devon_monitor": "google/gemini-2.5-flash", # Devon polling — just decisions, keep cheap
"background_gen": "x-ai/grok-4", # wilder, more specific caller backgrounds
"call_summary": "google/gemini-2.5-flash", # post-call summaries
"news_summary": "google/gemini-2.5-flash", # news digests
"topic_gen": "google/gemini-2.5-flash", # topic generation
"caller_dialog": "x-ai/grok-4.1-fast", # fallback if style_matched disabled ($0.20/$0.50)
"devon_ask": "x-ai/grok-4.1-fast", # Devon matches show energy, cheap ($0.20/$0.50)
"devon_monitor": "google/gemini-2.5-flash", # just yes/no decisions, keep cheap ($0.15/$0.60)
"background_gen": "x-ai/grok-4.1-fast", # wilder caller backgrounds ($0.20/$0.50)
"call_summary": "google/gemini-2.5-flash", # post-call, no personality needed ($0.15/$0.60)
"news_summary": "google/gemini-2.5-flash", # just digesting headlines ($0.15/$0.60)
"topic_gen": "google/gemini-2.5-flash", # structured output ($0.15/$0.60)
}
# TTS Settings
backend/main.py
+31 -25
View File
@@ -6240,34 +6240,40 @@ class Session:
# Caller model routing
self.caller_model_strategy: str = "style_matched" # "single" | "cycle" | "style_matched"
self.caller_model_pool: list[str] = [
"x-ai/grok-4",
"anthropic/claude-sonnet-4-5",
"mistralai/mistral-medium-3",
"qwen/qwen3-235b-a22b",
"deepseek/deepseek-chat-v3-0324",
"google/gemini-2.5-pro",
"meta-llama/llama-4-maverick",
"x-ai/grok-4.1-fast", # edgy, casual, great value ($0.20/$0.50)
"anthropic/claude-sonnet-4.6", # empathetic, nuanced ($3/$15)
"mistralai/mistral-large-2512", # dry wit, precise ($0.50/$1.50)
"deepseek/deepseek-r1-distill-llama-70b", # raw reasoning ($0.70/$0.80)
"meta-llama/llama-3.3-70b-instruct", # casual, natural ($0.10/$0.32)
"google/gemini-2.5-flash", # analytical ($0.30/$2.50)
]
self.caller_model_map: dict[str, str] = {
"high_energy": "x-ai/grok-4",
"confrontational": "x-ai/grok-4",
"angry_venting": "x-ai/grok-4",
"bragger": "x-ai/grok-4",
"comedian": "x-ai/grok-4",
"quiet_nervous": "anthropic/claude-sonnet-4-5",
"sweet_earnest": "anthropic/claude-sonnet-4-5",
"emotional": "anthropic/claude-sonnet-4-5",
"deadpan": "mistralai/mistral-medium-3",
"mysterious": "mistralai/mistral-medium-3",
"world_weary": "mistralai/mistral-medium-3",
"storyteller": "qwen/qwen3-235b-a22b",
"rambling": "qwen/qwen3-235b-a22b",
"oversharer": "deepseek/deepseek-chat-v3-0324",
"conspiracy": "deepseek/deepseek-chat-v3-0324",
"know_it_all": "google/gemini-2.5-pro",
"first_time": "meta-llama/llama-4-maverick",
# Grok 4.1 Fast — edgy, provocative, unfiltered humor
"high_energy": "x-ai/grok-4.1-fast",
"confrontational": "x-ai/grok-4.1-fast",
"angry_venting": "x-ai/grok-4.1-fast",
"bragger": "x-ai/grok-4.1-fast",
"comedian": "x-ai/grok-4.1-fast",
# Claude Sonnet 4.6 — empathetic, genuine emotional depth
"quiet_nervous": "anthropic/claude-sonnet-4.6",
"sweet_earnest": "anthropic/claude-sonnet-4.6",
"emotional": "anthropic/claude-sonnet-4.6",
# Mistral Large — dry, witty, precise delivery
"deadpan": "mistralai/mistral-large-2512",
"mysterious": "mistralai/mistral-large-2512",
"world_weary": "mistralai/mistral-large-2512",
# DeepSeek R1 Distill — raw, unfiltered, commits to the bit
"storyteller": "deepseek/deepseek-r1-distill-llama-70b",
"oversharer": "deepseek/deepseek-r1-distill-llama-70b",
"conspiracy": "deepseek/deepseek-r1-distill-llama-70b",
"rambling": "deepseek/deepseek-r1-distill-llama-70b",
# Gemini 2.5 Flash — articulate, analytical, cites facts
"know_it_all": "google/gemini-2.5-flash",
# Llama 3.3 70B — casual, natural hesitation, first-timer energy
"first_time": "meta-llama/llama-3.3-70b-instruct",
"reluctant_caller": "meta-llama/llama-3.3-70b-instruct",
}
self.caller_model_fallback: str = "anthropic/claude-sonnet-4-5"
self.caller_model_fallback: str = "anthropic/claude-sonnet-4.6"
self.caller_models: dict[str, str] = {} # caller_key → assigned model
self._caller_model_cycle_idx: int = 0
backend/services/cost_tracker.py
+23 -10
View File
@@ -32,25 +32,38 @@ class TTSCallRecord:
# OpenRouter pricing per 1M tokens (as of March 2026)
OPENROUTER_PRICING = {
# Claude
"anthropic/claude-sonnet-4.6": {"prompt": 3.00, "completion": 15.00},
"anthropic/claude-sonnet-4-5": {"prompt": 3.00, "completion": 15.00},
"anthropic/claude-haiku-4.5": {"prompt": 0.80, "completion": 4.00},
"anthropic/claude-3-haiku": {"prompt": 0.25, "completion": 1.25},
# Grok
"x-ai/grok-4.1-fast": {"prompt": 0.20, "completion": 0.50},
"x-ai/grok-4": {"prompt": 3.00, "completion": 15.00},
"x-ai/grok-4-fast": {"prompt": 5.00, "completion": 15.00},
"minimax/minimax-m2-her": {"prompt": 0.50, "completion": 1.50},
"mistralai/mistral-small-creative": {"prompt": 0.20, "completion": 0.60},
# Mistral
"mistralai/mistral-large-2512": {"prompt": 0.50, "completion": 1.50},
"mistralai/mistral-small-2603": {"prompt": 0.15, "completion": 0.60},
"mistralai/mistral-medium-3": {"prompt": 0.40, "completion": 2.00},
"mistralai/mistral-small-creative": {"prompt": 0.10, "completion": 0.30},
# DeepSeek
"deepseek/deepseek-r1-distill-llama-70b": {"prompt": 0.70, "completion": 0.80},
"deepseek/deepseek-chat-v3-0324": {"prompt": 0.27, "completion": 1.10},
"deepseek/deepseek-v3.2": {"prompt": 0.14, "completion": 0.28},
"google/gemini-2.5-flash": {"prompt": 0.15, "completion": 0.60},
# Google
"google/gemini-2.5-flash": {"prompt": 0.30, "completion": 2.50},
"google/gemini-2.5-pro": {"prompt": 1.25, "completion": 10.00},
"google/gemini-3-flash-preview": {"prompt": 0.50, "completion": 3.00},
"google/gemini-flash-1.5": {"prompt": 0.075, "completion": 0.30},
# Meta
"meta-llama/llama-3.3-70b-instruct": {"prompt": 0.10, "completion": 0.32},
"meta-llama/llama-4-maverick": {"prompt": 0.20, "completion": 0.60},
# Other
"moonshotai/kimi-k2": {"prompt": 0.60, "completion": 2.00},
"qwen/qwen3-235b-a22b": {"prompt": 0.20, "completion": 0.60},
"minimax/minimax-m2-her": {"prompt": 0.50, "completion": 1.50},
"openai/gpt-4o-mini": {"prompt": 0.15, "completion": 0.60},
"openai/gpt-4o": {"prompt": 2.50, "completion": 10.00},
"meta-llama/llama-3.1-8b-instruct": {"prompt": 0.06, "completion": 0.06},
"deepseek/deepseek-chat-v3-0324": {"prompt": 0.27, "completion": 1.10},
"moonshotai/kimi-k2": {"prompt": 0.60, "completion": 2.00},
"mistralai/mistral-medium-3": {"prompt": 0.40, "completion": 2.00},
"meta-llama/llama-4-maverick": {"prompt": 0.20, "completion": 0.60},
"qwen/qwen3-235b-a22b": {"prompt": 0.20, "completion": 0.60},
"google/gemini-2.5-pro": {"prompt": 1.25, "completion": 10.00},
}
# TTS pricing per character
backend/services/llm.py
+16 -16
View File
@@ -10,26 +10,26 @@ from .cost_tracker import cost_tracker
# Available OpenRouter models
OPENROUTER_MODELS = [
# Default
"anthropic/claude-sonnet-4-5",
# Best for natural dialog
# Primary
"anthropic/claude-sonnet-4.6",
"x-ai/grok-4.1-fast",
"x-ai/grok-4",
"x-ai/grok-4-fast",
"minimax/minimax-m2-her",
"mistralai/mistral-small-creative",
"deepseek/deepseek-v3.2",
# Other
"anthropic/claude-haiku-4.5",
# Style-matched pool
"mistralai/mistral-large-2512",
"deepseek/deepseek-r1-distill-llama-70b",
"meta-llama/llama-3.3-70b-instruct",
"google/gemini-2.5-flash",
"openai/gpt-4o-mini",
"openai/gpt-4o",
# New dialog models
# Other good options
"anthropic/claude-sonnet-4-5",
"anthropic/claude-haiku-4.5",
"deepseek/deepseek-chat-v3-0324",
"moonshotai/kimi-k2",
"mistralai/mistral-medium-3",
"meta-llama/llama-4-maverick",
"qwen/qwen3-235b-a22b",
"mistralai/mistral-small-2603",
"google/gemini-2.5-pro",
"google/gemini-3-flash-preview",
"x-ai/grok-4-fast",
"moonshotai/kimi-k2",
"qwen/qwen3-235b-a22b",
"meta-llama/llama-4-maverick",
# Legacy
"anthropic/claude-3-haiku",
"google/gemini-flash-1.5",