roadmodel 0.2.89__tar.gz → 0.2.91__tar.gz

{roadmodel-0.2.89 → roadmodel-0.2.91}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: roadmodel
-Version: 0.2.89
+Version: 0.2.91
 Summary: BYO-key CLI that recommends the right AI model, platform, and settings for a prompt.
 Project-URL: Homepage, https://roadmodel.ai
 Project-URL: Repository, https://github.com/nathanramoscfa/roadmodel

{roadmodel-0.2.89/roadmodel/data → roadmodel-0.2.91/docs/templates}/phase-roadmap-template.md

@@ -753,11 +753,15 @@ that owns it.}}
      **Runs:** line exactly as `roadmodel ladder --category
      <c> --complexity <x> [--novel]` prints them for the
      step's classification (MCP tool: `roadmap_settings`).
-     Rows: Model / Backup / Platform / Effort / Thinking /
-     Conversation. The Effort cell names each column's own
-     dial (`Intelligence Medium` on Codex, `Effort High` on
-     Claude Code); a column whose surface has no such
-     control reads `—` there. Every model is the latest
+     Rows: Model / Backup / Platform / Effort / AA Index /
+     Thinking / Conversation. The Effort cell names each
+     column's own dial (`Intelligence Medium` on Codex,
+     `Effort High` on Claude Code); a column whose surface
+     has no such control reads `—` there. The AA Index cell
+     is that column's model's Artificial Analysis
+     Intelligence Index at that column's effort, the figure
+     the pick was read from; it reads `not measured` where
+     AA has no row for the model at that effort. Every model is the latest
      version of its series, never an older one (GPT-6 Sol
      where GPT-6.1 Sol runs, Opus 5 where Opus 5.5 does).
      Set Conversation to **Continue** in every column only
@@ -869,6 +873,7 @@ reached}}):
 | Backup       | {{Backup — platform · dial}} | {{Backup — platform · dial}} | {{Backup — platform · dial}} |
 | Platform     | {{Access method}}           | {{Access method}}           | {{Access method}}           |
 | Effort       | {{Its dial and level}}      | {{Its dial and level}}      | {{Its dial and level}}      |
+| AA Index     | {{AA Index at that effort}} | {{AA Index at that effort}} | {{AA Index at that effort}} |
 | Thinking     | {{On, or — with no toggle}} | {{On, or — with no toggle}} | {{On, or — with no toggle}} |
 | Conversation | **{{New or Continue}}**     | **{{New or Continue}}**     | **{{New or Continue}}**     |
 
@@ -1290,6 +1295,7 @@ carrying ONLY the rows that platform's panel exposes.}}
 | Backup       | {{Backup — platform · dial}} | {{Backup — platform · dial}} | {{Backup — platform · dial}} |
 | Platform     | {{Access method}}           | {{Access method}}           | {{Access method}}           |
 | Effort       | {{Its dial and level}}      | {{Its dial and level}}      | {{Its dial and level}}      |
+| AA Index     | {{AA Index at that effort}} | {{AA Index at that effort}} | {{AA Index at that effort}} |
 | Thinking     | {{On, or — with no toggle}} | {{On, or — with no toggle}} | {{On, or — with no toggle}} |
 | Conversation | **{{New or Continue}}**     | **{{New or Continue}}**     | **{{New or Continue}}**     |
 
@@ -1426,6 +1432,7 @@ single-column variant that matches PLATFORM.}}
 | Backup       | {{Backup — platform · dial}} | {{Backup — platform · dial}} | {{Backup — platform · dial}} |
 | Platform     | {{Access method}}           | {{Access method}}           | {{Access method}}           |
 | Effort       | {{Its dial and level}}      | {{Its dial and level}}      | {{Its dial and level}}      |
+| AA Index     | {{AA Index at that effort}} | {{AA Index at that effort}} | {{AA Index at that effort}} |
 | Thinking     | {{On, or — with no toggle}} | {{On, or — with no toggle}} | {{On, or — with no toggle}} |
 | Conversation | **New**                     | **New**                     | **New**                     |
 

{roadmodel-0.2.89/roadmodel/data → roadmodel-0.2.91/docs/templates}/project-roadmap-template.md

@@ -254,6 +254,7 @@ dependency graph in §5 is honoured.
 | Backup       | <Backup — its platform · its dial> | <Backup — its platform · its dial> | <Backup — its platform · its dial> |
 | Platform     | <Access method>        | <Access method>        | <Claude Code, or the access method that pays> |
 | Effort       | <Its dial and level>   | <Its dial and level>   | <Deepest rung there — Effort Max in Claude Code> |
+| AA Index     | <AA Index at that effort> | <AA Index at that effort> | <AA Index at that effort>        |
 | Thinking     | <On, or — with no toggle> | <On, or — with no toggle> | <On, or — with no toggle>          |
 | Conversation | **New**                | **New**                | **New**                                  |
 
@@ -314,6 +315,7 @@ key pays for the platform.>
 | Backup       | …       | …        | …       |
 | Platform     | …       | …        | …       |
 | Effort       | …       | …        | …       |
+| AA Index     | …       | …        | …       |
 | Thinking     | …       | …        | …       |
 | Conversation | **New** | **New**  | **New** |
 
@@ -351,6 +353,7 @@ key pays for the platform.>
 | Backup       | …       | …        | …       |
 | Platform     | …       | …        | …       |
 | Effort       | …       | …        | …       |
+| AA Index     | …       | …        | …       |
 | Thinking     | …       | …        | …       |
 | Conversation | **New** | **New**  | **New** |
 

{roadmodel-0.2.89 → roadmodel-0.2.91}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "roadmodel"
-version = "0.2.89"
+version = "0.2.91"
 description = "BYO-key CLI that recommends the right AI model, platform, and settings for a prompt."
 readme = "README.md"
 requires-python = ">=3.11"

{roadmodel-0.2.89/docs/templates → roadmodel-0.2.91/roadmodel/data}/phase-roadmap-template.md

@@ -753,11 +753,15 @@ that owns it.}}
      **Runs:** line exactly as `roadmodel ladder --category
      <c> --complexity <x> [--novel]` prints them for the
      step's classification (MCP tool: `roadmap_settings`).
-     Rows: Model / Backup / Platform / Effort / Thinking /
-     Conversation. The Effort cell names each column's own
-     dial (`Intelligence Medium` on Codex, `Effort High` on
-     Claude Code); a column whose surface has no such
-     control reads `—` there. Every model is the latest
+     Rows: Model / Backup / Platform / Effort / AA Index /
+     Thinking / Conversation. The Effort cell names each
+     column's own dial (`Intelligence Medium` on Codex,
+     `Effort High` on Claude Code); a column whose surface
+     has no such control reads `—` there. The AA Index cell
+     is that column's model's Artificial Analysis
+     Intelligence Index at that column's effort, the figure
+     the pick was read from; it reads `not measured` where
+     AA has no row for the model at that effort. Every model is the latest
      version of its series, never an older one (GPT-6 Sol
      where GPT-6.1 Sol runs, Opus 5 where Opus 5.5 does).
      Set Conversation to **Continue** in every column only
@@ -869,6 +873,7 @@ reached}}):
 | Backup       | {{Backup — platform · dial}} | {{Backup — platform · dial}} | {{Backup — platform · dial}} |
 | Platform     | {{Access method}}           | {{Access method}}           | {{Access method}}           |
 | Effort       | {{Its dial and level}}      | {{Its dial and level}}      | {{Its dial and level}}      |
+| AA Index     | {{AA Index at that effort}} | {{AA Index at that effort}} | {{AA Index at that effort}} |
 | Thinking     | {{On, or — with no toggle}} | {{On, or — with no toggle}} | {{On, or — with no toggle}} |
 | Conversation | **{{New or Continue}}**     | **{{New or Continue}}**     | **{{New or Continue}}**     |
 
@@ -1290,6 +1295,7 @@ carrying ONLY the rows that platform's panel exposes.}}
 | Backup       | {{Backup — platform · dial}} | {{Backup — platform · dial}} | {{Backup — platform · dial}} |
 | Platform     | {{Access method}}           | {{Access method}}           | {{Access method}}           |
 | Effort       | {{Its dial and level}}      | {{Its dial and level}}      | {{Its dial and level}}      |
+| AA Index     | {{AA Index at that effort}} | {{AA Index at that effort}} | {{AA Index at that effort}} |
 | Thinking     | {{On, or — with no toggle}} | {{On, or — with no toggle}} | {{On, or — with no toggle}} |
 | Conversation | **{{New or Continue}}**     | **{{New or Continue}}**     | **{{New or Continue}}**     |
 
@@ -1426,6 +1432,7 @@ single-column variant that matches PLATFORM.}}
 | Backup       | {{Backup — platform · dial}} | {{Backup — platform · dial}} | {{Backup — platform · dial}} |
 | Platform     | {{Access method}}           | {{Access method}}           | {{Access method}}           |
 | Effort       | {{Its dial and level}}      | {{Its dial and level}}      | {{Its dial and level}}      |
+| AA Index     | {{AA Index at that effort}} | {{AA Index at that effort}} | {{AA Index at that effort}} |
 | Thinking     | {{On, or — with no toggle}} | {{On, or — with no toggle}} | {{On, or — with no toggle}} |
 | Conversation | **New**                     | **New**                     | **New**                     |
 

{roadmodel-0.2.89/docs/templates → roadmodel-0.2.91/roadmodel/data}/project-roadmap-template.md

@@ -254,6 +254,7 @@ dependency graph in §5 is honoured.
 | Backup       | <Backup — its platform · its dial> | <Backup — its platform · its dial> | <Backup — its platform · its dial> |
 | Platform     | <Access method>        | <Access method>        | <Claude Code, or the access method that pays> |
 | Effort       | <Its dial and level>   | <Its dial and level>   | <Deepest rung there — Effort Max in Claude Code> |
+| AA Index     | <AA Index at that effort> | <AA Index at that effort> | <AA Index at that effort>        |
 | Thinking     | <On, or — with no toggle> | <On, or — with no toggle> | <On, or — with no toggle>          |
 | Conversation | **New**                | **New**                | **New**                                  |
 
@@ -314,6 +315,7 @@ key pays for the platform.>
 | Backup       | …       | …        | …       |
 | Platform     | …       | …        | …       |
 | Effort       | …       | …        | …       |
+| AA Index     | …       | …        | …       |
 | Thinking     | …       | …        | …       |
 | Conversation | **New** | **New**  | **New** |
 
@@ -351,6 +353,7 @@ key pays for the platform.>
 | Backup       | …       | …        | …       |
 | Platform     | …       | …        | …       |
 | Effort       | …       | …        | …       |
+| AA Index     | …       | …        | …       |
 | Thinking     | …       | …        | …       |
 | Conversation | **New** | **New**  | **New** |
 

{roadmodel-0.2.89 → roadmodel-0.2.91}/src/roadmodel/__init__.py

@@ -1,2 +1,2 @@
 # src/roadmodel/__init__.py
-__version__ = "0.2.89"
+__version__ = "0.2.91"

{roadmodel-0.2.89 → roadmodel-0.2.91}/src/roadmodel/providers/anthropic.py

@@ -8,7 +8,15 @@ from roadmodel import usage
 from roadmodel.errors import ProviderCallError
 from roadmodel.providers._safety import hosted_timeout, private_transport, safe_error
 
-DEFAULT_MODEL = "claude-sonnet-4-6"
+# The package's own engine when no model is named (`roadmodel recommend`, the MCP
+# server, an SDK caller): one of service/app/engines.json's evaluated engines,
+# with the settings docs/engine-eval.json measured it at. The settings apply only
+# to that default; a named model runs with whatever the caller passes.
+# tests/test_engine_eval_mirror.py holds all three to the registry.
+# Sonnet 4.6 until 0.2.91; Sonnet 5.5 is that series' latest version.
+DEFAULT_MODEL = "claude-sonnet-5-5"
+DEFAULT_THINKING_BUDGET: int | None = 0
+DEFAULT_TEMPERATURE: float | None = None
 
 # claude-<family>-<major>[-<minor>][-<yyyymmdd>]: claude-haiku-4-5-20251001,
 # claude-sonnet-5, claude-opus-5-5, claude-fable-5-1.
@@ -81,6 +89,9 @@ def recommend(
     if missing_sdk is not None:
         raise missing_sdk
 
+    if model is None:
+        thinking_budget = DEFAULT_THINKING_BUDGET if thinking_budget is None else thinking_budget
+        temperature = DEFAULT_TEMPERATURE if temperature is None else temperature
     model_id = model or DEFAULT_MODEL
     # `kwargs` is typed `Any` for the same reason as providers/openai.py: the
     # SDK's overloaded `create` signature rejects a mixed-value dict literal.

{roadmodel-0.2.89 → roadmodel-0.2.91}/src/roadmodel/providers/google.py

@@ -8,14 +8,20 @@ from roadmodel import usage
 from roadmodel.errors import ProviderCallError
 from roadmodel.providers._safety import hosted_timeout, private_transport, safe_error
 
-# Default engine model for direct Google API calls when no model is passed.
-# Must be a model that is actually callable on the public Google Generative
-# Language API (v1beta generateContent). `gemini-3.1-pro` is catalog-only
-# (Cursor) and returns 404 NOT_FOUND on the direct API, so a bare
-# `roadmodel recommend` / the MCP server / any SDK caller with only a Google
-# key would fail; `gemini-2.5-pro` is the highest-quality callable Gemini.
-# Prod (the hosted service) forces an explicit model and never relies on this.
-DEFAULT_MODEL = "gemini-2.5-pro"
+# The package's own engine when no model is named (`roadmodel recommend`, the MCP
+# server, an SDK caller): one of service/app/engines.json's evaluated engines,
+# with the settings docs/engine-eval.json measured it at. The settings apply only
+# to that default; a named model runs with whatever the caller passes.
+# tests/test_engine_eval_mirror.py holds all three to the registry.
+# It must be callable by a NEW key: Google answers 404 NOT_FOUND to a project
+# that never used an older model, so `gemini-2.5-pro` (the default until
+# 0.2.90) and `gemini-3.1-pro` fail there while still working for keys that
+# used them before. Thinking level low and temperature 0 (#176) are how the
+# eval ran it (p50 2.5 s); left unset, the model thought at its default level
+# and a recommendation took about 30 s.
+DEFAULT_MODEL = "gemini-3.8-flash"
+DEFAULT_THINKING_BUDGET: int | None = 0
+DEFAULT_TEMPERATURE: float | None = 0.0
 
 # The 2.x generation takes a numeric thinking_budget; from Gemini 3 on, the API
 # documents discrete thinking LEVELS instead (update/gemini-thinking.json), so a
@@ -79,6 +85,9 @@ def recommend(
     if missing_sdk is not None:
         raise missing_sdk
 
+    if model is None:
+        thinking_budget = DEFAULT_THINKING_BUDGET if thinking_budget is None else thinking_budget
+        temperature = DEFAULT_TEMPERATURE if temperature is None else temperature
     try:
         client = genai.Client(
             api_key=api_key,

{roadmodel-0.2.89 → roadmodel-0.2.91}/src/roadmodel/providers/openai.py

@@ -8,7 +8,15 @@ from roadmodel import usage
 from roadmodel.errors import ProviderCallError
 from roadmodel.providers._safety import hosted_timeout, private_transport, safe_error
 
-DEFAULT_MODEL = "gpt-5.4"
+# The package's own engine when no model is named (`roadmodel recommend`, the MCP
+# server, an SDK caller): one of service/app/engines.json's evaluated engines,
+# with the settings docs/engine-eval.json measured it at. The settings apply only
+# to that default; a named model runs with whatever the caller passes.
+# tests/test_engine_eval_mirror.py holds all three to the registry.
+# gpt-5.4 until 0.2.91.
+DEFAULT_MODEL = "gpt-6-luna"
+DEFAULT_THINKING_BUDGET: int | None = 0
+DEFAULT_TEMPERATURE: float | None = None
 
 
 # The reasoning generations: gpt-5 and everything after it (gpt-6, gpt-6.1,
@@ -74,6 +82,9 @@ def recommend(
     # 32s, empty output). The recommender is a structured-classification task
     # where the pick is easy (T1), so a low effort is fast + cheap + sufficient;
     # 0 -> minimal for the highest-volume anon path.
+    if model is None:
+        thinking_budget = DEFAULT_THINKING_BUDGET if thinking_budget is None else thinking_budget
+        temperature = DEFAULT_TEMPERATURE if temperature is None else temperature
     _ = temperature
     missing_sdk: ProviderCallError | None = None
     try:

{roadmodel-0.2.89 → roadmodel-0.2.91}/src/roadmodel/scoring.py

@@ -2319,6 +2319,20 @@ def _thinking_cell(c: Candidate, catalog: dict[str, Any]) -> str:
     return "—" if model.get("always_thinks") else "On"
 
 
+def ladder_aa_index(bench: dict[str, Any], model_id: str, level: str) -> float | None:
+    """The AA Intelligence Index of ``model_id`` run at ``level``: AA's row at
+    that effort, or the headline row when AA names no level for the model (it
+    then stands for the model at whatever effort it runs). None when AA has
+    measured the model at other efforts only, or not at all."""
+    if measured_levels(bench, model_id):
+        return _evidence(bench, model_id, FRONTIER_INDEX, normalize_level(level))
+    return _evidence(bench, model_id, FRONTIER_INDEX)
+
+
+def _aa_cell(value: float | None) -> str:
+    return "not measured" if value is None else f"{value:g}"
+
+
 def _deepest_level(c: Candidate, catalog: dict[str, Any]) -> str:
     """The top rung of ``c``'s dial a roadmap may pin (``max``)."""
     levels = native_levels(c.platform_id, c.model_id, catalog=catalog) or list(EFFORT_LADDER)
@@ -2342,15 +2356,18 @@ def settings_ladder(
     so the table outlives today's pool states: the step's runner applies
     :data:`RUN_RULE` to the pools on the day instead of the roadmap being
     refreshed. ``phase`` is a phase-roadmap session: it runs Quality, at
-    the deepest rung its dial allows. Returns the columns and the Markdown
-    table; None when no pool model carries an AA Index."""
+    the deepest rung its dial allows. Each column carries its model's AA
+    Intelligence Index at the column's effort (:func:`ladder_aa_index`).
+    Returns the columns and the Markdown table; None when no pool model
+    carries an AA Index."""
     cat = catalog if catalog is not None else _cost._load_catalog()
+    bench = benchmarks if benchmarks is not None else _load_benchmarks()
     lad = ladder(
         task,
         user_context_text,
         unavailable_models=unavailable_models,
         catalog=cat,
-        benchmarks=benchmarks,
+        benchmarks=bench,
         neutral_pools=True,
     )
     if lad is None:
@@ -2369,6 +2386,7 @@ def settings_ladder(
             "platform": c.platform_name,
             "effort": normalize_level(level),
             "effort_cell": _effort_cell(c, level),
+            "aa_index": ladder_aa_index(bench, c.model_id, level),
             "thinking_cell": _thinking_cell(c, cat),
             "backup": None if rung.backup is None else backup_cell(rung.backup),
         }
@@ -2378,6 +2396,7 @@ def settings_ladder(
         ("Backup", [columns[t]["backup"] or "None" for t in LADDER_TIERS]),
         ("Platform", [columns[t]["platform"] for t in LADDER_TIERS]),
         ("Effort", [columns[t]["effort_cell"] for t in LADDER_TIERS]),
+        ("AA Index", [_aa_cell(columns[t]["aa_index"]) for t in LADDER_TIERS]),
         ("Thinking", [columns[t]["thinking_cell"] for t in LADDER_TIERS]),
         ("Conversation", ["**New**"] * len(LADDER_TIERS)),
     ]