deep-life-sci/models.yaml at main · langchain-samples/deep-life-sci · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
# Model and effort per role. Edits apply to the next message; no restart needed.

# Model ids: Anthropic models are bare (e.g. claude-sonnet-5); any other model needs a provider
# prefix (e.g. openai/gpt-5.6-terra)

# Effort must be a level the model supports; leave it empty to omit it
# This is required for models with no effort setting)

# Provider API keys go in LangSmith under Settings -> Integrations -> Provider Secrets

# Amazon Bedrock: prefix the Bedrock model or inference profile id with bedrock/
# (bedrock/openai.gpt-5.6-terra, bedrock/us.anthropic.claude-...) and add the
# AWS_BEARER_TOKEN_BEDROCK Provider Secret in LangSmith

# Main agent
root:
  model: openai/gpt-5.6-terra  # Anthropic: claude-sonnet-5
  effort: high                 # Anthropic: high

# Sub-agents spawned for tasks like reading papers
subagent:
  model: openai/gpt-5.6-luna   # Anthropic: claude-haiku-4-5-20251001
  effort: low                  # Anthropic: leave empty (Haiku 4.5 has no effort setting)

# Web search
# Note: web search on Bedrock needs one of its GPT models (gpt-5.4 or later); Claude has no web search there
search:
  model: openai/gpt-5.6-luna   # Anthropic: claude-haiku-4-5-20251001
  effort: low                  # Anthropic: leave empty (Haiku 4.5 has no effort setting)

# Judge for LLM-as-a-judge evals in LangSmith
judge:
  model: openai/gpt-5.6-terra  # Anthropic: claude-sonnet-5
  effort: low                  # Anthropic: low

labels:
  openai/gpt-5.6-terra: GPT-5.6 Terra
  openai/gpt-5.6-luna: GPT-5.6 Luna
  openai/gpt-5.6-sol: GPT-5.6 Sol
  claude-sonnet-5: Claude Sonnet 5
  claude-haiku-4-5-20251001: Claude Haiku 4.5