model_matrix.toml

schema_version = "1.0.0"
default_matrix = "current_balanced"

[matrices.current_balanced]
description = "Current balanced cross-provider core matrix."
models = [
  "gpt_5_6_sol",
  "gpt_5_6_luna",
  "claude_sonnet_5",
  "claude_haiku_4_5",
]

[matrices.discovery_current]
description = "Current two-provider matrix for discovery-track decisions."
models = [
  "gpt_5_6_luna",
  "claude_sonnet_5",
]

[matrices.current_optional]
description = "Optional Terra balanced-tier and Fable upper-bound models; not part of the default core."
models = [
  "gpt_5_6_terra",
  "claude_fable_5",
]

[matrices.current_full]
description = "Current balanced core plus optional comparison models."
models = [
  "gpt_5_6_sol",
  "gpt_5_6_luna",
  "claude_sonnet_5",
  "claude_haiku_4_5",
  "gpt_5_6_terra",
  "claude_fable_5",
]

[matrices.prior_modern]
description = "Superseded 2026 OpenAI comparison pair retained for reproducibility."
models = [
  "gpt_5_5_snapshot",
  "gpt_5_4_mini_snapshot",
]

[matrices.legacy_2026q2]
description = "Exact requested model aliases used by the frozen 2026 Q2 benchmark artifacts."
models = [
  "gpt_4o_mini_legacy",
  "gpt_4o_legacy",
  "claude_haiku_4_5_legacy",
  "claude_sonnet_4_5_legacy",
]

[models.gpt_5_6_sol]
inspect_id = "openai/gpt-5.6-sol"
provider = "openai"
expected_resolved_model = "gpt-5.6-sol"
display_name = "GPT-5.6 Sol"
short_name = "gpt-5.6-sol"
tier = "frontier"
color = "#168b5b"
aliases = ["openai/gpt-5.6"]

[models.gpt_5_6_sol.generate]
max_tokens = 16384
reasoning_effort = "medium"

[models.gpt_5_6_sol.inspect_model_info]
organization = "OpenAI"
model = "GPT-5.6 Sol"
knowledge_cutoff_date = "2026-02-16"
context_length = 1050000
output_tokens = 128000
reasoning = true
reasoning_effort_default = "medium"

[models.gpt_5_6_luna]
inspect_id = "openai/gpt-5.6-luna"
provider = "openai"
expected_resolved_model = "gpt-5.6-luna"
display_name = "GPT-5.6 Luna"
short_name = "gpt-5.6-luna"
tier = "efficient"
color = "#72c79a"

[models.gpt_5_6_luna.generate]
max_tokens = 16384
reasoning_effort = "medium"

[models.gpt_5_6_luna.inspect_model_info]
organization = "OpenAI"
model = "GPT-5.6 Luna"
knowledge_cutoff_date = "2026-02-16"
context_length = 1050000
output_tokens = 128000
reasoning = true
reasoning_effort_default = "medium"

[models.gpt_5_6_terra]
inspect_id = "openai/gpt-5.6-terra"
provider = "openai"
expected_resolved_model = "gpt-5.6-terra"
display_name = "GPT-5.6 Terra"
short_name = "gpt-5.6-terra"
tier = "balanced_optional"
color = "#075e3d"

[models.gpt_5_6_terra.generate]
max_tokens = 16384
reasoning_effort = "medium"

[models.gpt_5_6_terra.inspect_model_info]
organization = "OpenAI"
model = "GPT-5.6 Terra"
knowledge_cutoff_date = "2026-02-16"
context_length = 1050000
output_tokens = 128000
reasoning = true
reasoning_effort_default = "medium"

[models.claude_sonnet_5]
inspect_id = "anthropic/claude-sonnet-5"
provider = "anthropic"
expected_resolved_model = "claude-sonnet-5"
display_name = "Claude Sonnet 5"
short_name = "sonnet-5"
tier = "frontier"
color = "#c05621"

[models.claude_sonnet_5.generate]
max_tokens = 16384
reasoning_effort = "medium"

[models.claude_haiku_4_5]
inspect_id = "anthropic/claude-haiku-4-5-20251001"
provider = "anthropic"
expected_resolved_model = "claude-haiku-4-5-20251001"
display_name = "Claude Haiku 4.5"
short_name = "haiku-4.5"
tier = "efficient"
color = "#e09f7d"

[models.claude_haiku_4_5.generate]
max_tokens = 16384
reasoning_effort = "medium"

[models.claude_fable_5]
inspect_id = "anthropic/claude-fable-5"
provider = "anthropic"
expected_resolved_model = "claude-fable-5"
display_name = "Claude Fable 5"
short_name = "fable-5"
tier = "upper_bound"
color = "#8f341b"

[models.claude_fable_5.generate]
max_tokens = 16384
reasoning_effort = "medium"

[models.gpt_5_5_snapshot]
inspect_id = "openai/gpt-5.5-2026-04-23"
provider = "openai"
expected_resolved_model = "gpt-5.5-2026-04-23"
display_name = "GPT-5.5 (2026-04-23)"
short_name = "gpt-5.5"
tier = "prior_modern"
color = "#2e9f72"

[models.gpt_5_5_snapshot.generate]
max_tokens = 16384
reasoning_effort = "medium"

[models.gpt_5_4_mini_snapshot]
inspect_id = "openai/gpt-5.4-mini-2026-03-17"
provider = "openai"
expected_resolved_model = "gpt-5.4-mini-2026-03-17"
display_name = "GPT-5.4 mini (2026-03-17)"
short_name = "gpt-5.4-mini"
tier = "prior_modern"
color = "#89cfaa"

[models.gpt_5_4_mini_snapshot.generate]
max_tokens = 16384
reasoning_effort = "medium"

[models.gpt_4o_mini_legacy]
inspect_id = "openai/gpt-4o-mini"
provider = "openai"
expected_resolved_model = "gpt-4o-mini-2024-07-18"
display_name = "GPT-4o mini"
short_name = "gpt-4o-mini"
tier = "legacy"
color = "#7fcf9f"

[models.gpt_4o_mini_legacy.generate]
max_tokens = 4096
temperature = 0.0

[models.gpt_4o_legacy]
inspect_id = "openai/gpt-4o"
provider = "openai"
expected_resolved_model = "gpt-4o-2024-08-06"
display_name = "GPT-4o"
short_name = "gpt-4o"
tier = "legacy"
color = "#2ea572"

[models.gpt_4o_legacy.generate]
max_tokens = 4096
temperature = 0.0

[models.claude_haiku_4_5_legacy]
inspect_id = "anthropic/claude-haiku-4-5"
provider = "anthropic"
expected_resolved_model = "claude-haiku-4-5-20251001"
display_name = "Claude Haiku 4.5"
short_name = "haiku-4.5"
tier = "legacy"
color = "#e09f7d"

[models.claude_haiku_4_5_legacy.generate]
max_tokens = 4096
temperature = 0.0

[models.claude_sonnet_4_5_legacy]
inspect_id = "anthropic/claude-sonnet-4-5"
provider = "anthropic"
expected_resolved_model = "claude-sonnet-4-5-20250929"
display_name = "Claude Sonnet 4.5"
short_name = "sonnet-4.5"
tier = "legacy"
color = "#c05621"

[models.claude_sonnet_4_5_legacy.generate]
max_tokens = 4096
temperature = 0.0