roadmodel 0.2.67__tar.gz → 0.2.69__tar.gz

{roadmodel-0.2.67 → roadmodel-0.2.69}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: roadmodel
-Version: 0.2.67
+Version: 0.2.69
 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-0.2.67 → roadmodel-0.2.69}/docs/benchmarks.json

@@ -6,7 +6,7 @@
     "endpoint": "https://artificialanalysis.ai/api/v2/data/llms/models",
     "attribution": "Benchmark data © Artificial Analysis, used under its API terms with attribution."
   },
-  "generated_at_utc": "2026-10-05T22:11:00Z",
+  "generated_at_utc": "2026-10-06T22:11:12Z",
   "model_count": 54,
   "unmapped": [],
   "missing_slugs": [],
@@ -38,8 +38,8 @@
         "tau2": 0.546783625730994,
         "tau_banking": 0.0927835051546392
       },
-      "median_output_tokens_per_second": 107.429,
-      "median_time_to_first_token_seconds": 7.79
+      "median_output_tokens_per_second": 98.659,
+      "median_time_to_first_token_seconds": 7.086
     },
     "claude-fable-5": {
       "aa_id": "cd55210d-358e-4df1-ba9c-9acb5f186cc9",
@@ -96,8 +96,8 @@
         "tau2": null,
         "tau_banking": 0.472164948453608
       },
-      "median_output_tokens_per_second": 68.612,
-      "median_time_to_first_token_seconds": 145.612
+      "median_output_tokens_per_second": 69.251,
+      "median_time_to_first_token_seconds": 140.452
     },
     "claude-opus-5": {
       "aa_id": "b8fc61f7-5e9a-49e6-8547-6ac56db24627",
@@ -154,8 +154,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 96.593,
-      "median_time_to_first_token_seconds": 489.353
+      "median_output_tokens_per_second": 96.823,
+      "median_time_to_first_token_seconds": 470.118
     },
     "claude-sonnet-5": {
       "aa_id": "23c86e4a-c769-43c0-a056-79e3cd15834f",
@@ -212,7 +212,7 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 137.507,
+      "median_output_tokens_per_second": 137.452,
       "median_time_to_first_token_seconds": 269.765
     },
     "deepseek-flash": {
@@ -241,8 +241,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 214.176,
-      "median_time_to_first_token_seconds": 0.802
+      "median_output_tokens_per_second": 211.532,
+      "median_time_to_first_token_seconds": 0.838
     },
     "deepseek-v4-pro": {
       "aa_id": "83173329-c09a-41f1-a028-a282a5f908d7",
@@ -270,8 +270,8 @@
         "tau2": null,
         "tau_banking": 0.395876288659794
       },
-      "median_output_tokens_per_second": 103.076,
-      "median_time_to_first_token_seconds": 1.112
+      "median_output_tokens_per_second": 94.405,
+      "median_time_to_first_token_seconds": 1.116
     },
     "gemini-2.5-flash": {
       "aa_id": "219ed587-60c5-4a48-9517-8480e08d0ca1",
@@ -386,8 +386,8 @@
         "tau2": 0.956140350877193,
         "tau_banking": 0.214432989690722
       },
-      "median_output_tokens_per_second": 123.821,
-      "median_time_to_first_token_seconds": 15.32
+      "median_output_tokens_per_second": 120.916,
+      "median_time_to_first_token_seconds": 16.499
     },
     "gemini-3.5-flash": {
       "aa_id": "0097ebf5-124f-42f6-9463-33b00e711f03",
@@ -502,8 +502,8 @@
         "tau2": null,
         "tau_banking": 0.449484536082474
       },
-      "median_output_tokens_per_second": 223.213,
-      "median_time_to_first_token_seconds": 9.822
+      "median_output_tokens_per_second": 193.352,
+      "median_time_to_first_token_seconds": 11.478
     },
     "glm-4.5-air": {
       "aa_id": "5d303dc9-c027-401f-9803-4e9aa3331007",
@@ -618,8 +618,8 @@
         "tau2": null,
         "tau_banking": 0.503092783505155
       },
-      "median_output_tokens_per_second": 75.232,
-      "median_time_to_first_token_seconds": 3.1
+      "median_output_tokens_per_second": 74.56,
+      "median_time_to_first_token_seconds": 2.696
     },
     "glm-5.3-flash": {
       "aa_id": "19496b81-9f41-4214-a77a-1df803b3c5ae",
@@ -647,8 +647,8 @@
         "tau2": null,
         "tau_banking": 0.472164948453608
       },
-      "median_output_tokens_per_second": 49.967,
-      "median_time_to_first_token_seconds": 2.638
+      "median_output_tokens_per_second": 48.776,
+      "median_time_to_first_token_seconds": 2.773
     },
     "gpt-5": {
       "aa_id": "48e50f00-1fd1-4acc-b337-61078aa341e6",
@@ -850,8 +850,8 @@
         "tau2": 0.859649122807018,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 102.301,
-      "median_time_to_first_token_seconds": 34.347
+      "median_output_tokens_per_second": 94.067,
+      "median_time_to_first_token_seconds": 34.541
     },
     "gpt-5.4": {
       "aa_id": "a89c4b28-2d8c-456e-88ea-255fb51fd2b6",
@@ -1053,8 +1053,8 @@
         "tau2": 0.862573099415205,
         "tau_banking": 0.402061855670103
       },
-      "median_output_tokens_per_second": 116.052,
-      "median_time_to_first_token_seconds": 82.187
+      "median_output_tokens_per_second": 103.757,
+      "median_time_to_first_token_seconds": 83.831
     },
     "gpt-6-astra": {
       "aa_id": "2f339a97-9a0d-499a-9cb5-e0db665bfa25",
@@ -1082,8 +1082,8 @@
         "tau2": null,
         "tau_banking": 0.414432989690722
       },
-      "median_output_tokens_per_second": 64.764,
-      "median_time_to_first_token_seconds": 191.318
+      "median_output_tokens_per_second": 50.735,
+      "median_time_to_first_token_seconds": 205.366
     },
     "gpt-6-luna": {
       "aa_id": "6fb13851-40ce-4bda-ad3f-6b38e0c5daa7",
@@ -1111,8 +1111,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 135.654,
-      "median_time_to_first_token_seconds": 75.426
+      "median_output_tokens_per_second": 133.905,
+      "median_time_to_first_token_seconds": 68.659
     },
     "gpt-6-sol": {
       "aa_id": "c50ea08c-88c0-4eb7-85b8-27f2bbf6d527",
@@ -1169,8 +1169,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 60.455,
-      "median_time_to_first_token_seconds": 157.396
+      "median_output_tokens_per_second": 57.423,
+      "median_time_to_first_token_seconds": 179.839
     },
     "gpt-oss-120b": {
       "aa_id": "f0083258-8646-45b8-8082-7aaf6c2ea82a",
@@ -1198,8 +1198,8 @@
         "tau2": 0.657894736842105,
         "tau_banking": 0.127835051546392
       },
-      "median_output_tokens_per_second": 199.818,
-      "median_time_to_first_token_seconds": 0.501
+      "median_output_tokens_per_second": 186.012,
+      "median_time_to_first_token_seconds": 0.502
     },
     "gpt-oss-20b": {
       "aa_id": "36f73aaf-d38a-4b56-a2b3-d04d17186910",
@@ -1227,8 +1227,8 @@
         "tau2": 0.60233918128655,
         "tau_banking": 0.0701030927835052
       },
-      "median_output_tokens_per_second": 227.873,
-      "median_time_to_first_token_seconds": 0.427
+      "median_output_tokens_per_second": 231.256,
+      "median_time_to_first_token_seconds": 0.412
     },
     "grok-4.3": {
       "aa_id": "948892b5-db03-4118-a4a8-ccd51ed871ea",
@@ -1314,8 +1314,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 72.511,
-      "median_time_to_first_token_seconds": 9.66
+      "median_output_tokens_per_second": 69.431,
+      "median_time_to_first_token_seconds": 8.037
     },
     "kimi-k2.7-code": {
       "aa_id": "8d0cb231-7303-452c-9923-a9620b948475",
@@ -1343,8 +1343,8 @@
         "tau2": 0.900584795321637,
         "tau_banking": 0.202061855670103
       },
-      "median_output_tokens_per_second": 105.681,
-      "median_time_to_first_token_seconds": 1.142
+      "median_output_tokens_per_second": 83.243,
+      "median_time_to_first_token_seconds": 1.298
     },
     "kimi-k3": {
       "aa_id": "f7d2fc3e-1f7b-405f-818c-07952a4af78f",
@@ -1372,8 +1372,8 @@
         "tau2": null,
         "tau_banking": 0.45979381443299
       },
-      "median_output_tokens_per_second": 45.199,
-      "median_time_to_first_token_seconds": 2.455
+      "median_output_tokens_per_second": 45.073,
+      "median_time_to_first_token_seconds": 2.953
     },
     "mistral-large-3": {
       "aa_id": "4928e950-7f37-4475-b0dc-c5bad781a321",
@@ -1401,8 +1401,8 @@
         "tau2": 0.245614035087719,
         "tau_banking": 0.0577319587628866
       },
-      "median_output_tokens_per_second": 79.265,
-      "median_time_to_first_token_seconds": 0.692
+      "median_output_tokens_per_second": 79.072,
+      "median_time_to_first_token_seconds": 0.688
     },
     "mistral-medium-3.5": {
       "aa_id": "dd059b25-d82a-4ead-82a4-4adceaaec48b",
@@ -1430,8 +1430,8 @@
         "tau2": 0.941520467836257,
         "tau_banking": 0.150515463917526
       },
-      "median_output_tokens_per_second": 168.037,
-      "median_time_to_first_token_seconds": 0.6
+      "median_output_tokens_per_second": 162.625,
+      "median_time_to_first_token_seconds": 0.595
     },
     "mistral-small-4": {
       "aa_id": "3fd96175-4ef1-434c-8795-f873aec2abc1",
@@ -1459,8 +1459,8 @@
         "tau2": 0.412280701754386,
         "tau_banking": 0.0494845360824742
       },
-      "median_output_tokens_per_second": 171.598,
-      "median_time_to_first_token_seconds": 0.559
+      "median_output_tokens_per_second": 163.865,
+      "median_time_to_first_token_seconds": 0.542
     },
     "muse-spark-1.3": {
       "aa_id": "9999672c-b687-4026-9c15-5cef02ff53bb",
@@ -1488,8 +1488,8 @@
         "tau2": null,
         "tau_banking": 0.505154639175258
       },
-      "median_output_tokens_per_second": 122.652,
-      "median_time_to_first_token_seconds": 28.442
+      "median_output_tokens_per_second": 115.306,
+      "median_time_to_first_token_seconds": 26.027
     },
     "opus-4.7": {
       "aa_id": "e9a09db3-8fd6-41dd-ba2f-20e0a2bff7f2",

{roadmodel-0.2.67 → roadmodel-0.2.69}/docs/catalog.json

@@ -1,9 +1,9 @@
 {
   "schema_version": "4",
   "output_contract_version": 2,
-  "generated_at_utc": "2026-10-06T20:14:04Z",
+  "generated_at_utc": "2026-10-06T22:11:12Z",
   "source_doc_sha256": {
-    "model-selector.txt": "b7abeb92fb4d9b85b70a4544369a49de103d5695bf6dd77ea996dc4e85db0a31",
+    "model-selector.txt": "c18d9168548b6079ffe7e75cf891691b7ef3a2bfaa296dcd9a596f1e0a69e1ed",
     "model-tier-cost-scale.md": "2b3b0eb82237eb5cb5bf468a3fe971bcaf0160ad7d0bc700698d2c16640bc41d"
   },
   "models": [
@@ -24,7 +24,7 @@
         "speed": "S"
       },
       "jurisdiction": "us",
-      "headline_benchmarks": "AA Intelligence Index 16.9 (reasoning); Output Speed 122.5 tokens/s; AA-Omniscience -4.2; latency leader among Claude family",
+      "headline_benchmarks": "AA Intelligence Index 16.9 (reasoning); HLE 10.4%; SciCode 42.2; Terminal-Bench 4.0 0.0; AA-LCR 0.743; Output Speed 98.7 tokens/s; AA-Omniscience -4.2; latency leader among Claude family",
       "pricing_notes": "Hidden by default; Bedrock/Vertex: regional endpoints +10% surcharge; Cache: writes 1.25x, reads 0.1x",
       "best_for": "Speed-optimized lowest-cost Claude model, ideal for simple completions, high-volume repetitive tasks, and latency-sensitive workflows where a lightweight capable response matters more than deep reasoning",
       "superseded_by": null,
@@ -474,7 +474,7 @@
         "speed": "S"
       },
       "jurisdiction": "us",
-      "headline_benchmarks": "AA Intelligence Index 40.9 (high); HLE 47.8% (high); Terminal-Bench 2.1 87.6; Output Speed 237.7 tokens/s",
+      "headline_benchmarks": "AA Intelligence Index 40.9 (high); HLE 47.8%; SciCode 56.6; Terminal-Bench 4.0 19.7; AA-LCR 0.813; Terminal-Bench 2.1 87.6; Output Speed 193.4 tokens/s",
       "pricing_notes": "-",
       "best_for": "Frontier-class multimodal input, plus strong coding, planning, agentic work, long context and knowledge, with very fast output, at a low price. Its AA Intelligence Index of 40.9 ranks 18th of 54 measured models.",
       "superseded_by": null,

{roadmodel-0.2.67 → roadmodel-0.2.69}/docs/model-selector.txt

@@ -1252,9 +1252,9 @@
              tier-coding="A" tier-planning="A" tier-agentic="A"
              tier-multimodal="S" tier-long-context="A" tier-knowledge="A"
              tier-speed="S"
-             headline-benchmarks="AA Intelligence Index 40.9 (high); HLE 47.8% (high); Terminal-Bench 2.1 87.6; Output Speed 237.7 tokens/s"
+             headline-benchmarks="AA Intelligence Index 40.9 (high); HLE 47.8%; SciCode 56.6; Terminal-Bench 4.0 19.7; AA-LCR 0.813; Terminal-Bench 2.1 87.6; Output Speed 193.4 tokens/s"
              pricing-notes="-"
-             prose-generated="best-for"
+             prose-generated="best-for headline-benchmarks"
              best-for="Frontier-class multimodal input, plus strong coding, planning, agentic work, long context and knowledge, with very fast output, at a low price. Its AA Intelligence Index of 40.9 ranks 18th of 54 measured models." />
       <model id="gpt-5.6-luna" name="GPT-5.6 Luna"
              input-price-per-1m="$0.20" output-price-per-1m="$1.20"
@@ -1281,8 +1281,9 @@
              tier-coding="C" tier-planning="C" tier-agentic="D"
              tier-multimodal="B" tier-long-context="A" tier-knowledge="D"
              tier-speed="S"
-             headline-benchmarks="AA Intelligence Index 16.9 (reasoning); Output Speed 122.5 tokens/s; AA-Omniscience -4.2; latency leader among Claude family"
+             headline-benchmarks="AA Intelligence Index 16.9 (reasoning); HLE 10.4%; SciCode 42.2; Terminal-Bench 4.0 0.0; AA-LCR 0.743; Output Speed 98.7 tokens/s; AA-Omniscience -4.2; latency leader among Claude family"
              pricing-notes="Hidden by default; Bedrock/Vertex: regional endpoints +10% surcharge; Cache: writes 1.25x, reads 0.1x"
+             prose-generated="headline-benchmarks"
              best-for="Speed-optimized lowest-cost Claude model, ideal for simple completions, high-volume repetitive tasks, and latency-sensitive workflows where a lightweight capable response matters more than deep reasoning" />
       <model id="muse-spark-1.3" name="Muse Spark 1.3"
              input-price-per-1m="$1.25" output-price-per-1m="$4.25"

{roadmodel-0.2.67 → roadmodel-0.2.69}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "roadmodel"
-version = "0.2.67"
+version = "0.2.69"
 description = "BYO-key CLI that recommends the right AI model, platform, and settings for a prompt."
 readme = "README.md"
 requires-python = ">=3.11"

{roadmodel-0.2.67 → roadmodel-0.2.69}/roadmodel/data/benchmarks.json

@@ -6,7 +6,7 @@
     "endpoint": "https://artificialanalysis.ai/api/v2/data/llms/models",
     "attribution": "Benchmark data © Artificial Analysis, used under its API terms with attribution."
   },
-  "generated_at_utc": "2026-10-05T22:11:00Z",
+  "generated_at_utc": "2026-10-06T22:11:12Z",
   "model_count": 54,
   "unmapped": [],
   "missing_slugs": [],
@@ -38,8 +38,8 @@
         "tau2": 0.546783625730994,
         "tau_banking": 0.0927835051546392
       },
-      "median_output_tokens_per_second": 107.429,
-      "median_time_to_first_token_seconds": 7.79
+      "median_output_tokens_per_second": 98.659,
+      "median_time_to_first_token_seconds": 7.086
     },
     "claude-fable-5": {
       "aa_id": "cd55210d-358e-4df1-ba9c-9acb5f186cc9",
@@ -96,8 +96,8 @@
         "tau2": null,
         "tau_banking": 0.472164948453608
       },
-      "median_output_tokens_per_second": 68.612,
-      "median_time_to_first_token_seconds": 145.612
+      "median_output_tokens_per_second": 69.251,
+      "median_time_to_first_token_seconds": 140.452
     },
     "claude-opus-5": {
       "aa_id": "b8fc61f7-5e9a-49e6-8547-6ac56db24627",
@@ -154,8 +154,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 96.593,
-      "median_time_to_first_token_seconds": 489.353
+      "median_output_tokens_per_second": 96.823,
+      "median_time_to_first_token_seconds": 470.118
     },
     "claude-sonnet-5": {
       "aa_id": "23c86e4a-c769-43c0-a056-79e3cd15834f",
@@ -212,7 +212,7 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 137.507,
+      "median_output_tokens_per_second": 137.452,
       "median_time_to_first_token_seconds": 269.765
     },
     "deepseek-flash": {
@@ -241,8 +241,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 214.176,
-      "median_time_to_first_token_seconds": 0.802
+      "median_output_tokens_per_second": 211.532,
+      "median_time_to_first_token_seconds": 0.838
     },
     "deepseek-v4-pro": {
       "aa_id": "83173329-c09a-41f1-a028-a282a5f908d7",
@@ -270,8 +270,8 @@
         "tau2": null,
         "tau_banking": 0.395876288659794
       },
-      "median_output_tokens_per_second": 103.076,
-      "median_time_to_first_token_seconds": 1.112
+      "median_output_tokens_per_second": 94.405,
+      "median_time_to_first_token_seconds": 1.116
     },
     "gemini-2.5-flash": {
       "aa_id": "219ed587-60c5-4a48-9517-8480e08d0ca1",
@@ -386,8 +386,8 @@
         "tau2": 0.956140350877193,
         "tau_banking": 0.214432989690722
       },
-      "median_output_tokens_per_second": 123.821,
-      "median_time_to_first_token_seconds": 15.32
+      "median_output_tokens_per_second": 120.916,
+      "median_time_to_first_token_seconds": 16.499
     },
     "gemini-3.5-flash": {
       "aa_id": "0097ebf5-124f-42f6-9463-33b00e711f03",
@@ -502,8 +502,8 @@
         "tau2": null,
         "tau_banking": 0.449484536082474
       },
-      "median_output_tokens_per_second": 223.213,
-      "median_time_to_first_token_seconds": 9.822
+      "median_output_tokens_per_second": 193.352,
+      "median_time_to_first_token_seconds": 11.478
     },
     "glm-4.5-air": {
       "aa_id": "5d303dc9-c027-401f-9803-4e9aa3331007",
@@ -618,8 +618,8 @@
         "tau2": null,
         "tau_banking": 0.503092783505155
       },
-      "median_output_tokens_per_second": 75.232,
-      "median_time_to_first_token_seconds": 3.1
+      "median_output_tokens_per_second": 74.56,
+      "median_time_to_first_token_seconds": 2.696
     },
     "glm-5.3-flash": {
       "aa_id": "19496b81-9f41-4214-a77a-1df803b3c5ae",
@@ -647,8 +647,8 @@
         "tau2": null,
         "tau_banking": 0.472164948453608
       },
-      "median_output_tokens_per_second": 49.967,
-      "median_time_to_first_token_seconds": 2.638
+      "median_output_tokens_per_second": 48.776,
+      "median_time_to_first_token_seconds": 2.773
     },
     "gpt-5": {
       "aa_id": "48e50f00-1fd1-4acc-b337-61078aa341e6",
@@ -850,8 +850,8 @@
         "tau2": 0.859649122807018,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 102.301,
-      "median_time_to_first_token_seconds": 34.347
+      "median_output_tokens_per_second": 94.067,
+      "median_time_to_first_token_seconds": 34.541
     },
     "gpt-5.4": {
       "aa_id": "a89c4b28-2d8c-456e-88ea-255fb51fd2b6",
@@ -1053,8 +1053,8 @@
         "tau2": 0.862573099415205,
         "tau_banking": 0.402061855670103
       },
-      "median_output_tokens_per_second": 116.052,
-      "median_time_to_first_token_seconds": 82.187
+      "median_output_tokens_per_second": 103.757,
+      "median_time_to_first_token_seconds": 83.831
     },
     "gpt-6-astra": {
       "aa_id": "2f339a97-9a0d-499a-9cb5-e0db665bfa25",
@@ -1082,8 +1082,8 @@
         "tau2": null,
         "tau_banking": 0.414432989690722
       },
-      "median_output_tokens_per_second": 64.764,
-      "median_time_to_first_token_seconds": 191.318
+      "median_output_tokens_per_second": 50.735,
+      "median_time_to_first_token_seconds": 205.366
     },
     "gpt-6-luna": {
       "aa_id": "6fb13851-40ce-4bda-ad3f-6b38e0c5daa7",
@@ -1111,8 +1111,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 135.654,
-      "median_time_to_first_token_seconds": 75.426
+      "median_output_tokens_per_second": 133.905,
+      "median_time_to_first_token_seconds": 68.659
     },
     "gpt-6-sol": {
       "aa_id": "c50ea08c-88c0-4eb7-85b8-27f2bbf6d527",
@@ -1169,8 +1169,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 60.455,
-      "median_time_to_first_token_seconds": 157.396
+      "median_output_tokens_per_second": 57.423,
+      "median_time_to_first_token_seconds": 179.839
     },
     "gpt-oss-120b": {
       "aa_id": "f0083258-8646-45b8-8082-7aaf6c2ea82a",
@@ -1198,8 +1198,8 @@
         "tau2": 0.657894736842105,
         "tau_banking": 0.127835051546392
       },
-      "median_output_tokens_per_second": 199.818,
-      "median_time_to_first_token_seconds": 0.501
+      "median_output_tokens_per_second": 186.012,
+      "median_time_to_first_token_seconds": 0.502
     },
     "gpt-oss-20b": {
       "aa_id": "36f73aaf-d38a-4b56-a2b3-d04d17186910",
@@ -1227,8 +1227,8 @@
         "tau2": 0.60233918128655,
         "tau_banking": 0.0701030927835052
       },
-      "median_output_tokens_per_second": 227.873,
-      "median_time_to_first_token_seconds": 0.427
+      "median_output_tokens_per_second": 231.256,
+      "median_time_to_first_token_seconds": 0.412
     },
     "grok-4.3": {
       "aa_id": "948892b5-db03-4118-a4a8-ccd51ed871ea",
@@ -1314,8 +1314,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 72.511,
-      "median_time_to_first_token_seconds": 9.66
+      "median_output_tokens_per_second": 69.431,
+      "median_time_to_first_token_seconds": 8.037
     },
     "kimi-k2.7-code": {
       "aa_id": "8d0cb231-7303-452c-9923-a9620b948475",
@@ -1343,8 +1343,8 @@
         "tau2": 0.900584795321637,
         "tau_banking": 0.202061855670103
       },
-      "median_output_tokens_per_second": 105.681,
-      "median_time_to_first_token_seconds": 1.142
+      "median_output_tokens_per_second": 83.243,
+      "median_time_to_first_token_seconds": 1.298
     },
     "kimi-k3": {
       "aa_id": "f7d2fc3e-1f7b-405f-818c-07952a4af78f",
@@ -1372,8 +1372,8 @@
         "tau2": null,
         "tau_banking": 0.45979381443299
       },
-      "median_output_tokens_per_second": 45.199,
-      "median_time_to_first_token_seconds": 2.455
+      "median_output_tokens_per_second": 45.073,
+      "median_time_to_first_token_seconds": 2.953
     },
     "mistral-large-3": {
       "aa_id": "4928e950-7f37-4475-b0dc-c5bad781a321",
@@ -1401,8 +1401,8 @@
         "tau2": 0.245614035087719,
         "tau_banking": 0.0577319587628866
       },
-      "median_output_tokens_per_second": 79.265,
-      "median_time_to_first_token_seconds": 0.692
+      "median_output_tokens_per_second": 79.072,
+      "median_time_to_first_token_seconds": 0.688
     },
     "mistral-medium-3.5": {
       "aa_id": "dd059b25-d82a-4ead-82a4-4adceaaec48b",
@@ -1430,8 +1430,8 @@
         "tau2": 0.941520467836257,
         "tau_banking": 0.150515463917526
       },
-      "median_output_tokens_per_second": 168.037,
-      "median_time_to_first_token_seconds": 0.6
+      "median_output_tokens_per_second": 162.625,
+      "median_time_to_first_token_seconds": 0.595
     },
     "mistral-small-4": {
       "aa_id": "3fd96175-4ef1-434c-8795-f873aec2abc1",
@@ -1459,8 +1459,8 @@
         "tau2": 0.412280701754386,
         "tau_banking": 0.0494845360824742
       },
-      "median_output_tokens_per_second": 171.598,
-      "median_time_to_first_token_seconds": 0.559
+      "median_output_tokens_per_second": 163.865,
+      "median_time_to_first_token_seconds": 0.542
     },
     "muse-spark-1.3": {
       "aa_id": "9999672c-b687-4026-9c15-5cef02ff53bb",
@@ -1488,8 +1488,8 @@
         "tau2": null,
         "tau_banking": 0.505154639175258
       },
-      "median_output_tokens_per_second": 122.652,
-      "median_time_to_first_token_seconds": 28.442
+      "median_output_tokens_per_second": 115.306,
+      "median_time_to_first_token_seconds": 26.027
     },
     "opus-4.7": {
       "aa_id": "e9a09db3-8fd6-41dd-ba2f-20e0a2bff7f2",

{roadmodel-0.2.67 → roadmodel-0.2.69}/roadmodel/data/catalog.json

@@ -1,9 +1,9 @@
 {
   "schema_version": "4",
   "output_contract_version": 2,
-  "generated_at_utc": "2026-10-06T20:14:04Z",
+  "generated_at_utc": "2026-10-06T22:11:12Z",
   "source_doc_sha256": {
-    "model-selector.txt": "b7abeb92fb4d9b85b70a4544369a49de103d5695bf6dd77ea996dc4e85db0a31",
+    "model-selector.txt": "c18d9168548b6079ffe7e75cf891691b7ef3a2bfaa296dcd9a596f1e0a69e1ed",
     "model-tier-cost-scale.md": "2b3b0eb82237eb5cb5bf468a3fe971bcaf0160ad7d0bc700698d2c16640bc41d"
   },
   "models": [
@@ -24,7 +24,7 @@
         "speed": "S"
       },
       "jurisdiction": "us",
-      "headline_benchmarks": "AA Intelligence Index 16.9 (reasoning); Output Speed 122.5 tokens/s; AA-Omniscience -4.2; latency leader among Claude family",
+      "headline_benchmarks": "AA Intelligence Index 16.9 (reasoning); HLE 10.4%; SciCode 42.2; Terminal-Bench 4.0 0.0; AA-LCR 0.743; Output Speed 98.7 tokens/s; AA-Omniscience -4.2; latency leader among Claude family",
       "pricing_notes": "Hidden by default; Bedrock/Vertex: regional endpoints +10% surcharge; Cache: writes 1.25x, reads 0.1x",
       "best_for": "Speed-optimized lowest-cost Claude model, ideal for simple completions, high-volume repetitive tasks, and latency-sensitive workflows where a lightweight capable response matters more than deep reasoning",
       "superseded_by": null,
@@ -474,7 +474,7 @@
         "speed": "S"
       },
       "jurisdiction": "us",
-      "headline_benchmarks": "AA Intelligence Index 40.9 (high); HLE 47.8% (high); Terminal-Bench 2.1 87.6; Output Speed 237.7 tokens/s",
+      "headline_benchmarks": "AA Intelligence Index 40.9 (high); HLE 47.8%; SciCode 56.6; Terminal-Bench 4.0 19.7; AA-LCR 0.813; Terminal-Bench 2.1 87.6; Output Speed 193.4 tokens/s",
       "pricing_notes": "-",
       "best_for": "Frontier-class multimodal input, plus strong coding, planning, agentic work, long context and knowledge, with very fast output, at a low price. Its AA Intelligence Index of 40.9 ranks 18th of 54 measured models.",
       "superseded_by": null,

{roadmodel-0.2.67 → roadmodel-0.2.69}/roadmodel/data/model-selector.txt

@@ -1252,9 +1252,9 @@
              tier-coding="A" tier-planning="A" tier-agentic="A"
              tier-multimodal="S" tier-long-context="A" tier-knowledge="A"
              tier-speed="S"
-             headline-benchmarks="AA Intelligence Index 40.9 (high); HLE 47.8% (high); Terminal-Bench 2.1 87.6; Output Speed 237.7 tokens/s"
+             headline-benchmarks="AA Intelligence Index 40.9 (high); HLE 47.8%; SciCode 56.6; Terminal-Bench 4.0 19.7; AA-LCR 0.813; Terminal-Bench 2.1 87.6; Output Speed 193.4 tokens/s"
              pricing-notes="-"
-             prose-generated="best-for"
+             prose-generated="best-for headline-benchmarks"
              best-for="Frontier-class multimodal input, plus strong coding, planning, agentic work, long context and knowledge, with very fast output, at a low price. Its AA Intelligence Index of 40.9 ranks 18th of 54 measured models." />
       <model id="gpt-5.6-luna" name="GPT-5.6 Luna"
              input-price-per-1m="$0.20" output-price-per-1m="$1.20"
@@ -1281,8 +1281,9 @@
              tier-coding="C" tier-planning="C" tier-agentic="D"
              tier-multimodal="B" tier-long-context="A" tier-knowledge="D"
              tier-speed="S"
-             headline-benchmarks="AA Intelligence Index 16.9 (reasoning); Output Speed 122.5 tokens/s; AA-Omniscience -4.2; latency leader among Claude family"
+             headline-benchmarks="AA Intelligence Index 16.9 (reasoning); HLE 10.4%; SciCode 42.2; Terminal-Bench 4.0 0.0; AA-LCR 0.743; Output Speed 98.7 tokens/s; AA-Omniscience -4.2; latency leader among Claude family"
              pricing-notes="Hidden by default; Bedrock/Vertex: regional endpoints +10% surcharge; Cache: writes 1.25x, reads 0.1x"
+             prose-generated="headline-benchmarks"
              best-for="Speed-optimized lowest-cost Claude model, ideal for simple completions, high-volume repetitive tasks, and latency-sensitive workflows where a lightweight capable response matters more than deep reasoning" />
       <model id="muse-spark-1.3" name="Muse Spark 1.3"
              input-price-per-1m="$1.25" output-price-per-1m="$4.25"

{roadmodel-0.2.67 → roadmodel-0.2.69}/scripts/updater_lock.py

@@ -1,9 +1,9 @@
 """One advisory updater lock per machine, with explicit child handover.
 
 The installed launcher holds the operating-system lock while a synchronously
-invoked replacement updater runs. Its direct child can identify that ownership
-through a random inherited token and owner PID. No project or rule data is
-stored in the persistent, harmless lock file.
+invoked replacement updater runs. Its child can identify that ownership through
+a random inherited token and owner PID, including Windows venv redirectors
+between the processes. No project or rule data is stored in the lock file.
 """
 
 from __future__ import annotations
@@ -52,18 +52,79 @@ def _release(stream: BinaryIO) -> None:
         fcntl.flock(stream.fileno(), fcntl.LOCK_UN)
 
 
+def _windows_process_parents() -> dict[int, int]:
+    """Return a native process snapshot without starting another interpreter.
+
+    A Windows venv's python.exe is a redirector which launches the base Python
+    as its child. The updater owner therefore need not be os.getppid().
+    """
+    import ctypes
+    from ctypes import wintypes
+
+    class ProcessEntry(ctypes.Structure):
+        _fields_ = [
+            ("dwSize", wintypes.DWORD),
+            ("cntUsage", wintypes.DWORD),
+            ("th32ProcessID", wintypes.DWORD),
+            ("th32DefaultHeapID", ctypes.c_size_t),
+            ("th32ModuleID", wintypes.DWORD),
+            ("cntThreads", wintypes.DWORD),
+            ("th32ParentProcessID", wintypes.DWORD),
+            ("pcPriClassBase", wintypes.LONG),
+            ("dwFlags", wintypes.DWORD),
+            ("szExeFile", wintypes.WCHAR * 260),
+        ]
+
+    kernel = ctypes.WinDLL("kernel32", use_last_error=True)
+    kernel.CreateToolhelp32Snapshot.argtypes = [wintypes.DWORD, wintypes.DWORD]
+    kernel.CreateToolhelp32Snapshot.restype = wintypes.HANDLE
+    for name in ("Process32FirstW", "Process32NextW"):
+        function = getattr(kernel, name)
+        function.argtypes = [wintypes.HANDLE, ctypes.POINTER(ProcessEntry)]
+        function.restype = wintypes.BOOL
+    kernel.CloseHandle.argtypes = [wintypes.HANDLE]
+    kernel.CloseHandle.restype = wintypes.BOOL
+    snapshot = kernel.CreateToolhelp32Snapshot(0x00000002, 0)  # TH32CS_SNAPPROCESS
+    if snapshot == ctypes.c_void_p(-1).value:
+        raise ctypes.WinError(ctypes.get_last_error())
+    try:
+        entry = ProcessEntry()
+        entry.dwSize = ctypes.sizeof(entry)
+        parents: dict[int, int] = {}
+        found = kernel.Process32FirstW(snapshot, ctypes.byref(entry))
+        while found:
+            parents[entry.th32ProcessID] = entry.th32ParentProcessID
+            found = kernel.Process32NextW(snapshot, ctypes.byref(entry))
+        return parents
+    finally:
+        kernel.CloseHandle(snapshot)
+
+
+def _owner_is_ancestor(pid: int) -> bool:
+    if pid == os.getpid():
+        return True
+    if not _WINDOWS:
+        return pid == os.getppid()
+    parents = _windows_process_parents()
+    current = os.getpid()
+    seen: set[int] = set()
+    while current in parents and current not in seen:
+        if current == pid:
+            return True
+        seen.add(current)
+        current = parents[current]
+    return False
+
+
 def _inherited_owner(stream: BinaryIO) -> bool:
     token = os.environ.get(LOCK_TOKEN_ENV)
     owner_pid = os.environ.get(LOCK_PID_ENV)
     if not token or not owner_pid:
         return False
     try:
-        # Handover is deliberately limited to a direct child (or reentrant
-        # use by the owner). A stale token in an unrelated shell is not proof
-        # that the lock holder will remain alive throughout this operation.
+        # Match the record before checking ancestry. A token copied into an
+        # unrelated shell is insufficient, as is a PID for a departed owner.
         pid = int(owner_pid)
-        if pid not in (os.getpid(), os.getppid()):
-            return False
         stream.seek(1)  # Windows locks byte zero; owner metadata is after it.
         state = json.loads(stream.read().decode("utf-8"))
         return (
@@ -73,6 +134,7 @@ def _inherited_owner(stream: BinaryIO) -> bool:
             and state["pid"] == pid
             and isinstance(state.get("token"), str)
             and secrets.compare_digest(state["token"], token)
+            and _owner_is_ancestor(pid)
         )
     except (OSError, ValueError, TypeError):
         return False

{roadmodel-0.2.67 → roadmodel-0.2.69}/src/roadmodel/__init__.py

@@ -1,2 +1,2 @@
 # src/roadmodel/__init__.py
-__version__ = "0.2.67"
+__version__ = "0.2.69"

{roadmodel-0.2.67 → roadmodel-0.2.69}/src/roadmodel/scoring.py

@@ -913,6 +913,10 @@ def rank(
             pj = str(method.get("provider_jurisdiction", "us")).lower()
             if pj != "local" and pj not in juris:
                 continue  # `local` runs on the operator's hardware: passes every list
+            if task.category == "speed" and str(method.get("billing", "")) == "local":
+                # Speed evidence is the hosted providers' measured throughput;
+                # the operator's own hardware runs at its own, unmeasured pace.
+                continue
             funding, state, scarcity, note = _funding_for(
                 _cost._with_model(method, model_id),
                 cat,
@@ -1210,9 +1214,12 @@ def ladder(
       category stands alone.
     - QUALITY is the adequate point that scores highest in the category (then
       the higher AA Index, then the cheaper). COST is the cheapest adequate
-      point. BALANCED is the point strictly between them with the best score
-      at the balanced posture; with none between, the COST or QUALITY model at
-      the balanced posture's effort, whichever differs from both other rungs.
+      point the operator has already paid for (a subscription with headroom,
+      or local weights); a per-token point takes COST only when no prepaid
+      point is adequate. BALANCED is the point strictly between them, stronger
+      than COST in the task's category, with the best score at the balanced
+      posture; with none, the COST or QUALITY model at the balanced posture's
+      effort, whichever differs from both other rungs.
     - Each rung runs at :func:`effort_for` its posture (cheap / balanced /
       best), set on its platform's own dial (:func:`native_level`) where the
       catalog documents one. A lower rung on the same model and platform as
@@ -1265,24 +1272,34 @@ def ladder(
     def same(a: Rung, b: Rung) -> bool:
         return a.candidate.model_id == b.candidate.model_id and a.level == b.level
 
-    cost_rung = rung("cost", span[0])
+    # COST is the cheapest adequate point already paid for: a subscription pool
+    # with headroom or local weights (scarcity below list price) costs the
+    # operator nothing new, while a per-token call is fresh spend.
+    cost_point = next((p for p in span if p.candidate.scarcity < 1.0), span[0])
+    lo = span.index(cost_point)
+    cost_rung = rung("cost", cost_point)
     quality_rung = (
         replace(rung("quality", specialist), specialist=True)
         if specialist is not None
         else rung("quality", top)
     )
     # With a specialist on QUALITY, BALANCED spans every adequate frontier point
-    # above COST, the frontier's strongest included.
-    between = span[1:] if specialist is not None else span[1:-1]
+    # above COST, the frontier's strongest included. A BALANCED point earns its
+    # higher price only by beating COST in the task's own category: the
+    # frontier is drawn on the AA Index, and a point above COST on that axis
+    # can still trail it in the category the task needs.
+    between = span[lo + 1 :] if specialist is not None else span[lo + 1 : -1]
+    between = [p for p in between if p.candidate.quality > cost_point.candidate.quality]
     if between:
         balanced_rung = rung(
             "balanced", max(between, key=lambda p: (p.candidate.score, -p.price_usd))
         )
     else:
-        # Two adequate points or one: BALANCED runs one of them at the balanced
-        # posture, whichever differs from both other rungs; it converges only
-        # when neither does. Beside a specialist it stays on the frontier.
-        options = [rung("balanced", span[0])]
+        # No point between COST and QUALITY beats COST in the category:
+        # BALANCED runs one of those two at the balanced posture, whichever
+        # differs from both other rungs; it converges only when neither does.
+        # Beside a specialist it stays on the frontier.
+        options = [rung("balanced", cost_point)]
         if specialist is None:
             options.append(rung("balanced", top))
         balanced_rung = next(