roadmodel 0.2.54__tar.gz → 0.2.55__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (49) hide show
  1. {roadmodel-0.2.54 → roadmodel-0.2.55}/PKG-INFO +1 -1
  2. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/catalog.json +38 -19
  3. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/model-selector.txt +12 -4
  4. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/model-tier-cost-scale.md +4 -4
  5. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/user-context.example.md +18 -13
  6. {roadmodel-0.2.54 → roadmodel-0.2.55}/pyproject.toml +1 -1
  7. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/catalog.json +38 -19
  8. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/model-selector.txt +12 -4
  9. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/model-tier-cost-scale.md +4 -4
  10. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/user-context.example.md +18 -13
  11. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/__init__.py +1 -1
  12. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/cost.py +14 -1
  13. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/scoring.py +11 -3
  14. {roadmodel-0.2.54 → roadmodel-0.2.55}/.gitignore +0 -0
  15. {roadmodel-0.2.54 → roadmodel-0.2.55}/LICENSE +0 -0
  16. {roadmodel-0.2.54 → roadmodel-0.2.55}/NOTICE +0 -0
  17. {roadmodel-0.2.54 → roadmodel-0.2.55}/README.md +0 -0
  18. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/benchmarks.json +0 -0
  19. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/settings-display.md +0 -0
  20. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/templates/phase-roadmap-template.md +0 -0
  21. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/templates/planning-kit-how-to-use.md +0 -0
  22. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/templates/project-roadmap-template.md +0 -0
  23. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/templates/prompt-phase-roadmap.md +0 -0
  24. {roadmodel-0.2.54 → roadmodel-0.2.55}/docs/templates/prompt-project-roadmap.md +0 -0
  25. {roadmodel-0.2.54 → roadmodel-0.2.55}/hatch_build.py +0 -0
  26. {roadmodel-0.2.54 → roadmodel-0.2.55}/infra/README.md +0 -0
  27. {roadmodel-0.2.54 → roadmodel-0.2.55}/infra/gcp-killswitch/README.md +0 -0
  28. {roadmodel-0.2.54 → roadmodel-0.2.55}/infra/supabase/README.md +0 -0
  29. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/benchmarks.json +0 -0
  30. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/phase-roadmap-template.md +0 -0
  31. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/planning-kit-how-to-use.md +0 -0
  32. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/project-roadmap-template.md +0 -0
  33. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/prompt-phase-roadmap.md +0 -0
  34. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/prompt-project-roadmap.md +0 -0
  35. {roadmodel-0.2.54 → roadmodel-0.2.55}/roadmodel/data/settings-display.md +0 -0
  36. {roadmodel-0.2.54 → roadmodel-0.2.55}/scripts/update_projects.py +0 -0
  37. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/__main__.py +0 -0
  38. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/cli.py +0 -0
  39. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/config.py +0 -0
  40. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/errors.py +0 -0
  41. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/mcp_server.py +0 -0
  42. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/providers/__init__.py +0 -0
  43. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/providers/anthropic.py +0 -0
  44. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/providers/google.py +0 -0
  45. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/providers/openai.py +0 -0
  46. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/providers/openai_compatible.py +0 -0
  47. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/providers/registry.py +0 -0
  48. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/recommend.py +0 -0
  49. {roadmodel-0.2.54 → roadmodel-0.2.55}/src/roadmodel/user_context.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: roadmodel
3
- Version: 0.2.54
3
+ Version: 0.2.55
4
4
  Summary: BYO-key CLI that recommends the right AI model, platform, and settings for a prompt.
5
5
  Project-URL: Homepage, https://roadmodel.ai
6
6
  Project-URL: Repository, https://github.com/nathanramoscfa/roadmodel
@@ -1,10 +1,10 @@
1
1
  {
2
2
  "schema_version": "3",
3
3
  "output_contract_version": 2,
4
- "generated_at_utc": "2026-10-02T00:26:36Z",
4
+ "generated_at_utc": "2026-10-02T02:50:29Z",
5
5
  "source_doc_sha256": {
6
- "model-selector.txt": "0c7966b41eedc3c8a5f38873d0b56b85575ee935dcef47d8b094d624ab5e34b6",
7
- "model-tier-cost-scale.md": "0e07bd4fae45e3472d064067a09f426c4430da87b6031ce7d610baeb23b5b2fb"
6
+ "model-selector.txt": "64998c7c21759d3b13f71908e5c51e3db4c95e1bae39efd5a3e391381b01a572",
7
+ "model-tier-cost-scale.md": "2b3b0eb82237eb5cb5bf468a3fe971bcaf0160ad7d0bc700698d2c16640bc41d"
8
8
  },
9
9
  "models": [
10
10
  {
@@ -1426,11 +1426,11 @@
1426
1426
  "jurisdiction": "us",
1427
1427
  "headline_benchmarks": "AA Intelligence Index 30.1 (adaptive reasoning, max effort); HLE 33.6%; SciCode 50.1; Terminal-Bench 4.0 3.0; AA-LCR 0.800; LMArena Text Elo 1457.7 (#41); AA-Omniscience 12.4",
1428
1428
  "pricing_notes": "Hidden by default; Requires Max Mode on legacy request-based plans; Up to 1M tokens with extended context at the same per-token rates (no long-context surcharge)",
1429
- "best_for": "Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models. Superseded by Sonnet 5.",
1430
- "superseded_by": "claude-sonnet-5",
1431
- "superseded_on": "2026-09-24",
1429
+ "best_for": "Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models.",
1430
+ "superseded_by": null,
1431
+ "superseded_on": null,
1432
1432
  "added_on": null,
1433
- "retires_on": "2026-10-24"
1433
+ "retires_on": null
1434
1434
  }
1435
1435
  ],
1436
1436
  "access_methods": [
@@ -1458,6 +1458,26 @@
1458
1458
  "exposes_orchestration": "no",
1459
1459
  "best_for": "Programmatic / scripted Claude use outside Claude Code \u2014 raw API headers, batch endpoints, or features not surfaced by Claude Code. Falls back here when claude.ai Max budget is exhausted."
1460
1460
  },
1461
+ {
1462
+ "id": "antigravity",
1463
+ "name": "Antigravity",
1464
+ "provider": "google",
1465
+ "provider_jurisdiction": "us",
1466
+ "billing": "subscription-included",
1467
+ "requires": "google-ai-pro-or-ultra-subscription",
1468
+ "supports_models": [
1469
+ "gemini-3.1-pro",
1470
+ "gemini-3.6-flash",
1471
+ "gemini-3.7-flash",
1472
+ "gemini-3.8-flash",
1473
+ "gpt-oss-120b",
1474
+ "sonnet-4.6"
1475
+ ],
1476
+ "exposes_max_mode": "no",
1477
+ "exposes_thinking": "yes",
1478
+ "exposes_orchestration": "no",
1479
+ "best_for": "Google's agent-first coding surface: the Antigravity desktop IDE, its VS Code panel and the `agy` CLI, which replaced the Gemini CLI for personal Google accounts on 2026-06-18. Google AI Pro and Ultra pay for it: quota refreshes every five hours up to a weekly limit (the free tier refreshes weekly), and both can buy AI credits for overage. It serves the Gemini models and, on Google's quota, Claude Sonnet 4.6 and Claude Opus 4.6 with thinking and gpt-oss-120b (antigravity.google/docs/models); Claude Opus 4.6 is not in this catalog. Gemini models take a Low / Medium / High reasoning level, set in the model id (`<model>-low` / `-medium` / `-high`) or with `agy --effort`; the Claude and GPT-OSS models carry their own thinking setting. It bills a Google AI plan or a Google Cloud project, not a Gemini API key."
1480
+ },
1461
1481
  {
1462
1482
  "id": "chatgpt-app",
1463
1483
  "name": "ChatGPT (web / desktop)",
@@ -1685,8 +1705,8 @@
1685
1705
  "name": "Gemini CLI",
1686
1706
  "provider": "google",
1687
1707
  "provider_jurisdiction": "us",
1688
- "billing": "subscription-or-key",
1689
- "requires": "gemini-advanced-subscription OR google-api-key",
1708
+ "billing": "per-token",
1709
+ "requires": "google-api-key",
1690
1710
  "supports_models": [
1691
1711
  "gemini-2.5-flash",
1692
1712
  "gemini-3-flash",
@@ -1700,7 +1720,7 @@
1700
1720
  "exposes_max_mode": "no",
1701
1721
  "exposes_thinking": "yes",
1702
1722
  "exposes_orchestration": "no",
1703
- "best_for": "Terminal-driven Gemini use; the CLI surface for multimodal and long-context Gemini work outside Cursor's pool. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) \u2014 per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash)."
1723
+ "best_for": "Since 2026-06-18 the Gemini CLI serves only paid Gemini API keys and Gemini Code Assist Standard / Enterprise licences: Google AI Pro, Ultra and free personal accounts moved to Antigravity (`antigravity`). Terminal-driven Gemini use; the CLI surface for multimodal and long-context Gemini work outside Cursor's pool. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) \u2014 per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash)."
1704
1724
  },
1705
1725
  {
1706
1726
  "id": "google-api",
@@ -1991,8 +2011,7 @@
1991
2011
  "monthly_usd": 4.99,
1992
2012
  "annual_usd": null,
1993
2013
  "surface_funded": [
1994
- "gemini-app",
1995
- "gemini-cli"
2014
+ "gemini-app"
1996
2015
  ],
1997
2016
  "notes": "Entry-paid Google AI tier with 2x higher usage limits than Free in the Gemini app, access to Gemini 3.1 Pro / Nano Banana Pro / Daily Brief / Gemini Omni video generation, 200 Google Flow Credits, and 400 GB of cloud storage (price cut from $7.99 to $4.99 on 2026-06-08; storage doubled from 200 GB to 400 GB)."
1998
2017
  },
@@ -2002,10 +2021,10 @@
2002
2021
  "monthly_usd": 19.99,
2003
2022
  "annual_usd": 199.99,
2004
2023
  "surface_funded": [
2005
- "gemini-app",
2006
- "gemini-cli"
2024
+ "antigravity",
2025
+ "gemini-app"
2007
2026
  ],
2008
- "notes": "Gemini 3.1 Pro and supporting multimodal features in the Gemini app and Gemini CLI, plus Deep Research, Nano Banana Pro, Veo 3.1 access, 1,000 monthly AI credits, and 5 TB of Google One storage; includes YouTube Premium Lite; 50% off the first year for new subscribers."
2027
+ "notes": "Gemini 3.1 Pro and supporting multimodal features in the Gemini app, Antigravity quota refreshed every five hours up to a weekly limit, plus Deep Research, Nano Banana Pro, Veo 3.1 access, 1,000 monthly AI credits, and 5 TB of Google One storage; includes YouTube Premium Lite; 50% off the first year for new subscribers."
2009
2028
  },
2010
2029
  {
2011
2030
  "provider": "Google",
@@ -2013,8 +2032,8 @@
2013
2032
  "monthly_usd": 99.99,
2014
2033
  "annual_usd": null,
2015
2034
  "surface_funded": [
2016
- "gemini-app",
2017
- "gemini-cli"
2035
+ "antigravity",
2036
+ "gemini-app"
2018
2037
  ],
2019
2038
  "notes": "Developer-focused Ultra tier (May 2026): 5x Pro usage limits in Gemini app and Google Antigravity, Gemini 3.5 Flash integration, priority access to Antigravity, 20 TB cloud storage, and YouTube Premium individual plan."
2020
2039
  },
@@ -2024,8 +2043,8 @@
2024
2043
  "monthly_usd": 199.99,
2025
2044
  "annual_usd": null,
2026
2045
  "surface_funded": [
2027
- "gemini-app",
2028
- "gemini-cli"
2046
+ "antigravity",
2047
+ "gemini-app"
2029
2048
  ],
2030
2049
  "notes": "Highest access to Gemini 3.1 Pro, Deep Think, Project Genie, Veo 3.1 with 25,000 monthly AI credits, $100/month Google Cloud credits, and 30 TB Google One storage (price reduced from $249.99 to $199.99 at Google I/O 2026)."
2031
2050
  },
@@ -1037,7 +1037,7 @@
1037
1037
  headline-benchmarks="AA Intelligence Index 30.1 (adaptive reasoning, max effort); HLE 33.6%; SciCode 50.1; Terminal-Bench 4.0 3.0; AA-LCR 0.800; LMArena Text Elo 1457.7 (#41); AA-Omniscience 12.4"
1038
1038
  pricing-notes="Hidden by default; Requires Max Mode on legacy request-based plans; Up to 1M tokens with extended context at the same per-token rates (no long-context surcharge)"
1039
1039
  prose-generated="best-for headline-benchmarks"
1040
- best-for="Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models. Superseded by Sonnet 5." superseded-by="claude-sonnet-5" superseded-on="2026-09-24"/>
1040
+ best-for="Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models."/>
1041
1041
 
1042
1042
  <model id="gpt-5.4" name="GPT-5.4"
1043
1043
  input-price-per-1m="$2.50" output-price-per-1m="$15.00"
@@ -1601,13 +1601,13 @@
1601
1601
  exposes-orchestration="no"
1602
1602
  best-for="Programmatic / scripted Gemini use with a Google API key. Pay-per-token at Google's published rates. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) shared across the 3.x and 2.5 generations — per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash). Powers the roadmodel SaaS free-tier surfaces (/recommend on Gemini 2.5 Flash; /roadmap on Gemini 2.5 Flash with 3.1 Pro escalation)." />
1603
1603
  <method id="gemini-cli" name="Gemini CLI"
1604
- provider="google" billing="subscription-or-key"
1604
+ provider="google" billing="per-token"
1605
1605
  provider-jurisdiction="us"
1606
- requires="gemini-advanced-subscription OR google-api-key"
1606
+ requires="google-api-key"
1607
1607
  supports-models="gemini-3.1-pro,gemini-3-pro,gemini-3.8-flash,gemini-3.7-flash,gemini-3.6-flash,gemini-3.5-flash,gemini-3-flash,gemini-2.5-flash"
1608
1608
  exposes-max-mode="no" exposes-thinking="yes"
1609
1609
  exposes-orchestration="no"
1610
- best-for="Terminal-driven Gemini use; the CLI surface for multimodal and long-context Gemini work outside Cursor's pool. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) — per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash)." />
1610
+ best-for="Since 2026-06-18 the Gemini CLI serves only paid Gemini API keys and Gemini Code Assist Standard / Enterprise licences: Google AI Pro, Ultra and free personal accounts moved to Antigravity (`antigravity`). Terminal-driven Gemini use; the CLI surface for multimodal and long-context Gemini work outside Cursor's pool. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) — per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash)." />
1611
1611
  <method id="gemini-app" name="Gemini (web / app)"
1612
1612
  provider="google" billing="subscription-included"
1613
1613
  provider-jurisdiction="us"
@@ -1616,6 +1616,14 @@
1616
1616
  exposes-max-mode="no" exposes-thinking="yes"
1617
1617
  exposes-orchestration="no"
1618
1618
  best-for="Chat-driven Gemini use under the Gemini Advanced subscription budget." />
1619
+ <method id="antigravity" name="Antigravity"
1620
+ provider="google" billing="subscription-included"
1621
+ provider-jurisdiction="us"
1622
+ requires="google-ai-pro-or-ultra-subscription"
1623
+ supports-models="gemini-3.1-pro,gemini-3.8-flash,gemini-3.7-flash,gemini-3.6-flash,sonnet-4.6,gpt-oss-120b"
1624
+ exposes-max-mode="no" exposes-thinking="yes"
1625
+ exposes-orchestration="no"
1626
+ best-for="Google's agent-first coding surface: the Antigravity desktop IDE, its VS Code panel and the `agy` CLI, which replaced the Gemini CLI for personal Google accounts on 2026-06-18. Google AI Pro and Ultra pay for it: quota refreshes every five hours up to a weekly limit (the free tier refreshes weekly), and both can buy AI credits for overage. It serves the Gemini models and, on Google's quota, Claude Sonnet 4.6 and Claude Opus 4.6 with thinking and gpt-oss-120b (antigravity.google/docs/models); Claude Opus 4.6 is not in this catalog. Gemini models take a Low / Medium / High reasoning level, set in the model id (`<model>-low` / `-medium` / `-high`) or with `agy --effort`; the Claude and GPT-OSS models carry their own thinking setting. It bills a Google AI plan or a Google Cloud project, not a Gemini API key." />
1619
1627
  <method id="xai-api" name="xAI API"
1620
1628
  provider="xai" billing="per-token"
1621
1629
  provider-jurisdiction="us"
@@ -207,10 +207,10 @@ and adjust the rebuild rules in [`update/prompt.md`](../update/prompt.md).
207
207
  | ChatGPT Pro ($100) | $100 | — | OpenAI | chatgpt-app, codex-cli | Same model suite as Pro $200 (GPT-5.6 Sol, GPT-5.5, o1 Pro mode) at 5x Plus usage limits; Codex access included with promotional 10x multiplier through May 31, 2026. |
208
208
  | ChatGPT Pro ($200) | $200 | — | OpenAI | chatgpt-app, codex-cli | Same core Pro features as Pro $100 (Pro models, Codex, deep research, image creation, memory, file uploads) with more usage than Pro $100; new subscriptions are open again, and new subscriptions not eligible for grandfathering get a lower usage allowance than the previous Pro 200. |
209
209
  | ChatGPT Pro ($500) | $500 | — | OpenAI | chatgpt-app, codex-cli | Highest-usage ChatGPT Pro tier (more included usage than Pro $100 and Pro $200); includes Astra Ultrafast plus the core Pro features (Pro models, Codex, deep research, image creation, memory, file uploads). |
210
- | Google AI Plus | $4.99 | — | Google | gemini-app, gemini-cli | Entry-paid Google AI tier with 2x higher usage limits than Free in the Gemini app, access to Gemini 3.1 Pro / Nano Banana Pro / Daily Brief / Gemini Omni video generation, 200 Google Flow Credits, and 400 GB of cloud storage (price cut from $7.99 to $4.99 on 2026-06-08; storage doubled from 200 GB to 400 GB). |
211
- | Google AI Pro | $19.99 | $199.99 | Google | gemini-app, gemini-cli | Gemini 3.1 Pro and supporting multimodal features in the Gemini app and Gemini CLI, plus Deep Research, Nano Banana Pro, Veo 3.1 access, 1,000 monthly AI credits, and 5 TB of Google One storage; includes YouTube Premium Lite; 50% off the first year for new subscribers. |
212
- | Google AI Ultra ($100) | $99.99 | — | Google | gemini-app, gemini-cli | Developer-focused Ultra tier (May 2026): 5x Pro usage limits in Gemini app and Google Antigravity, Gemini 3.5 Flash integration, priority access to Antigravity, 20 TB cloud storage, and YouTube Premium individual plan. |
213
- | Google AI Ultra ($200) | $199.99 | — | Google | gemini-app, gemini-cli | Highest access to Gemini 3.1 Pro, Deep Think, Project Genie, Veo 3.1 with 25,000 monthly AI credits, $100/month Google Cloud credits, and 30 TB Google One storage (price reduced from $249.99 to $199.99 at Google I/O 2026). |
210
+ | Google AI Plus | $4.99 | — | Google | gemini-app | Entry-paid Google AI tier with 2x higher usage limits than Free in the Gemini app, access to Gemini 3.1 Pro / Nano Banana Pro / Daily Brief / Gemini Omni video generation, 200 Google Flow Credits, and 400 GB of cloud storage (price cut from $7.99 to $4.99 on 2026-06-08; storage doubled from 200 GB to 400 GB). |
211
+ | Google AI Pro | $19.99 | $199.99 | Google | gemini-app, antigravity | Gemini 3.1 Pro and supporting multimodal features in the Gemini app, Antigravity quota refreshed every five hours up to a weekly limit, plus Deep Research, Nano Banana Pro, Veo 3.1 access, 1,000 monthly AI credits, and 5 TB of Google One storage; includes YouTube Premium Lite; 50% off the first year for new subscribers. |
212
+ | Google AI Ultra ($100) | $99.99 | — | Google | gemini-app, antigravity | Developer-focused Ultra tier (May 2026): 5x Pro usage limits in Gemini app and Google Antigravity, Gemini 3.5 Flash integration, priority access to Antigravity, 20 TB cloud storage, and YouTube Premium individual plan. |
213
+ | Google AI Ultra ($200) | $199.99 | — | Google | gemini-app, antigravity | Highest access to Gemini 3.1 Pro, Deep Think, Project Genie, Veo 3.1 with 25,000 monthly AI credits, $100/month Google Cloud credits, and 30 TB Google One storage (price reduced from $249.99 to $199.99 at Google I/O 2026). |
214
214
  | Cursor Pro | $20 | $192 | Cursor | cursor | Two monthly usage pools — Cursor Models (generous included usage for Grok 4.7, Grok 4.6, Grok 4.5, Composer 2.5) and Other Models (third-party models at API price); on-demand overage billed at API rates. |
215
215
  | Cursor Pro+ | $60 | $576 | Cursor | cursor | Same model coverage as Pro at roughly 3x the OpenAI / Claude / Gemini usage budget. |
216
216
  | Cursor Ultra | $200 | $1920 | Cursor | cursor | Same model coverage as Pro at roughly 20x the OpenAI / Claude / Gemini usage budget; priority access to new features. |
@@ -15,6 +15,13 @@ per-token prices) so that:
15
15
  The weekly catalog refresh in [`update/prompt.md`](../update/prompt.md)
16
16
  **must not** modify this file.
17
17
 
18
+ **Describe access, not models.** This file says what you pay for and how you
19
+ work. Leave model versions out: the catalog decides which models qualify and
20
+ keeps up with every release, so a version written here goes stale the day its
21
+ successor ships. Name a model family (Opus, Sonnet) only to state a
22
+ preference; the selector reads it as that family's newest model the catalog
23
+ does not mark superseded.
24
+
18
25
  ## How the selector consumes this file
19
26
 
20
27
  When [`model-selector.txt`](model-selector.txt) is referenced, the AI
@@ -186,8 +193,9 @@ dial the chosen platform lacks gets NO line at all — never `Off`, never
186
193
  - **Thinking** — default `On` wherever extended reasoning helps; `Off`
187
194
  only for purely mechanical steps where latency matters. It never
188
195
  carries an effort word or a number; `THINKING: Max` is not a setting
189
- any surface can apply. Claude Fable 5 cannot disable extended
190
- thinking, so never declare `Off` for it.
196
+ any surface can apply. Some models cannot disable extended
197
+ thinking (the Claude Fable line, for one), so never declare `Off`
198
+ for those.
191
199
  - **Codex / OpenAI** — the surface labels its dial "Intelligence" in its
192
200
  own UI, but the emitted field is still `EFFORT` (same value, different
193
201
  name). It exposes no separate thinking toggle and no Max Mode.
@@ -204,8 +212,8 @@ dial the chosen platform lacks gets NO line at all — never `Off`, never
204
212
  than a step inside one. By default each phase's **Phase roadmap
205
213
  settings** are picked quality first: the strongest planning model you
206
214
  can reach, at its deepest reasoning rung (`Max` in Claude Code). State
207
- a rule here to override that, for example "Phase roadmaps: Opus 5.5 at
208
- `XHigh`" to keep more of a weekly pool.
215
+ a rule here to override that, for example "Phase roadmaps: one rung below
216
+ the deepest" to keep more of a weekly pool.
209
217
 
210
218
  ## Budget priority and speed posture
211
219
 
@@ -232,14 +240,11 @@ the frontier tier on every posture.
232
240
  **Speed posture:** speed is NOT a valued dimension — workflows are
233
241
  batch / asynchronous, not latency-sensitive. Apply these rules:
234
242
 
235
- - **Never recommend a "Fast" variant** of a model (e.g., Opus 4.7
236
- Fast, GPT-5.4 Fast, GPT-5 Fast, Claude 4.6 Opus Fast mode) when a
237
- standard non-fast variant of the same base model is available.
238
- Fast variants in Cursor's catalog typically charge 2x output for
239
- marginal speed gain (e.g., "Fast mode is 15% faster with 2x
240
- pricing", "Claude 4.6 Opus Fast mode" at $30/$150 input/output vs.
241
- standard Opus at $5/$25) — the speed-for-cost trade is never worth
242
- it here.
243
+ - **Never recommend a "Fast" variant** of a model (e.g., an Opus Fast
244
+ or a GPT Fast mode) when a standard non-fast variant of the same base
245
+ model is available. Fast variants typically charge 2x or more on
246
+ output for a marginal speed gain — the speed-for-cost trade is never
247
+ worth it here.
243
248
  - **In `<selection-algorithm>` Step 5** (cost tie-breaker), when the
244
249
  tied set contains both a standard and a Fast variant of the same
245
250
  base model, the standard variant always wins regardless of any
@@ -358,7 +363,7 @@ platforms.excluded: (none declared)
358
363
  Values are **access-method ids** from the `<access-methods>` block in
359
364
  `model-selector.txt` — e.g. `claude-code`, `cursor`, `codex-cli`,
360
365
  `chatgpt-app`, `claude-web`, `anthropic-api`, `openai-api`,
361
- `google-api`, `gemini-cli`, `deepseek-api`, `mistral-api` — not
366
+ `google-api`, `gemini-cli`, `antigravity`, `deepseek-api`, `mistral-api` — not
362
367
  display names ("Claude Code") and not provider names ("Anthropic").
363
368
  An operator who works only in the terminal might declare:
364
369
 
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
4
4
 
5
5
  [project]
6
6
  name = "roadmodel"
7
- version = "0.2.54"
7
+ version = "0.2.55"
8
8
  description = "BYO-key CLI that recommends the right AI model, platform, and settings for a prompt."
9
9
  readme = "README.md"
10
10
  requires-python = ">=3.11"
@@ -1,10 +1,10 @@
1
1
  {
2
2
  "schema_version": "3",
3
3
  "output_contract_version": 2,
4
- "generated_at_utc": "2026-10-02T00:26:36Z",
4
+ "generated_at_utc": "2026-10-02T02:50:29Z",
5
5
  "source_doc_sha256": {
6
- "model-selector.txt": "0c7966b41eedc3c8a5f38873d0b56b85575ee935dcef47d8b094d624ab5e34b6",
7
- "model-tier-cost-scale.md": "0e07bd4fae45e3472d064067a09f426c4430da87b6031ce7d610baeb23b5b2fb"
6
+ "model-selector.txt": "64998c7c21759d3b13f71908e5c51e3db4c95e1bae39efd5a3e391381b01a572",
7
+ "model-tier-cost-scale.md": "2b3b0eb82237eb5cb5bf468a3fe971bcaf0160ad7d0bc700698d2c16640bc41d"
8
8
  },
9
9
  "models": [
10
10
  {
@@ -1426,11 +1426,11 @@
1426
1426
  "jurisdiction": "us",
1427
1427
  "headline_benchmarks": "AA Intelligence Index 30.1 (adaptive reasoning, max effort); HLE 33.6%; SciCode 50.1; Terminal-Bench 4.0 3.0; AA-LCR 0.800; LMArena Text Elo 1457.7 (#41); AA-Omniscience 12.4",
1428
1428
  "pricing_notes": "Hidden by default; Requires Max Mode on legacy request-based plans; Up to 1M tokens with extended context at the same per-token rates (no long-context surcharge)",
1429
- "best_for": "Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models. Superseded by Sonnet 5.",
1430
- "superseded_by": "claude-sonnet-5",
1431
- "superseded_on": "2026-09-24",
1429
+ "best_for": "Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models.",
1430
+ "superseded_by": null,
1431
+ "superseded_on": null,
1432
1432
  "added_on": null,
1433
- "retires_on": "2026-10-24"
1433
+ "retires_on": null
1434
1434
  }
1435
1435
  ],
1436
1436
  "access_methods": [
@@ -1458,6 +1458,26 @@
1458
1458
  "exposes_orchestration": "no",
1459
1459
  "best_for": "Programmatic / scripted Claude use outside Claude Code \u2014 raw API headers, batch endpoints, or features not surfaced by Claude Code. Falls back here when claude.ai Max budget is exhausted."
1460
1460
  },
1461
+ {
1462
+ "id": "antigravity",
1463
+ "name": "Antigravity",
1464
+ "provider": "google",
1465
+ "provider_jurisdiction": "us",
1466
+ "billing": "subscription-included",
1467
+ "requires": "google-ai-pro-or-ultra-subscription",
1468
+ "supports_models": [
1469
+ "gemini-3.1-pro",
1470
+ "gemini-3.6-flash",
1471
+ "gemini-3.7-flash",
1472
+ "gemini-3.8-flash",
1473
+ "gpt-oss-120b",
1474
+ "sonnet-4.6"
1475
+ ],
1476
+ "exposes_max_mode": "no",
1477
+ "exposes_thinking": "yes",
1478
+ "exposes_orchestration": "no",
1479
+ "best_for": "Google's agent-first coding surface: the Antigravity desktop IDE, its VS Code panel and the `agy` CLI, which replaced the Gemini CLI for personal Google accounts on 2026-06-18. Google AI Pro and Ultra pay for it: quota refreshes every five hours up to a weekly limit (the free tier refreshes weekly), and both can buy AI credits for overage. It serves the Gemini models and, on Google's quota, Claude Sonnet 4.6 and Claude Opus 4.6 with thinking and gpt-oss-120b (antigravity.google/docs/models); Claude Opus 4.6 is not in this catalog. Gemini models take a Low / Medium / High reasoning level, set in the model id (`<model>-low` / `-medium` / `-high`) or with `agy --effort`; the Claude and GPT-OSS models carry their own thinking setting. It bills a Google AI plan or a Google Cloud project, not a Gemini API key."
1480
+ },
1461
1481
  {
1462
1482
  "id": "chatgpt-app",
1463
1483
  "name": "ChatGPT (web / desktop)",
@@ -1685,8 +1705,8 @@
1685
1705
  "name": "Gemini CLI",
1686
1706
  "provider": "google",
1687
1707
  "provider_jurisdiction": "us",
1688
- "billing": "subscription-or-key",
1689
- "requires": "gemini-advanced-subscription OR google-api-key",
1708
+ "billing": "per-token",
1709
+ "requires": "google-api-key",
1690
1710
  "supports_models": [
1691
1711
  "gemini-2.5-flash",
1692
1712
  "gemini-3-flash",
@@ -1700,7 +1720,7 @@
1700
1720
  "exposes_max_mode": "no",
1701
1721
  "exposes_thinking": "yes",
1702
1722
  "exposes_orchestration": "no",
1703
- "best_for": "Terminal-driven Gemini use; the CLI surface for multimodal and long-context Gemini work outside Cursor's pool. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) \u2014 per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash)."
1723
+ "best_for": "Since 2026-06-18 the Gemini CLI serves only paid Gemini API keys and Gemini Code Assist Standard / Enterprise licences: Google AI Pro, Ultra and free personal accounts moved to Antigravity (`antigravity`). Terminal-driven Gemini use; the CLI surface for multimodal and long-context Gemini work outside Cursor's pool. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) \u2014 per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash)."
1704
1724
  },
1705
1725
  {
1706
1726
  "id": "google-api",
@@ -1991,8 +2011,7 @@
1991
2011
  "monthly_usd": 4.99,
1992
2012
  "annual_usd": null,
1993
2013
  "surface_funded": [
1994
- "gemini-app",
1995
- "gemini-cli"
2014
+ "gemini-app"
1996
2015
  ],
1997
2016
  "notes": "Entry-paid Google AI tier with 2x higher usage limits than Free in the Gemini app, access to Gemini 3.1 Pro / Nano Banana Pro / Daily Brief / Gemini Omni video generation, 200 Google Flow Credits, and 400 GB of cloud storage (price cut from $7.99 to $4.99 on 2026-06-08; storage doubled from 200 GB to 400 GB)."
1998
2017
  },
@@ -2002,10 +2021,10 @@
2002
2021
  "monthly_usd": 19.99,
2003
2022
  "annual_usd": 199.99,
2004
2023
  "surface_funded": [
2005
- "gemini-app",
2006
- "gemini-cli"
2024
+ "antigravity",
2025
+ "gemini-app"
2007
2026
  ],
2008
- "notes": "Gemini 3.1 Pro and supporting multimodal features in the Gemini app and Gemini CLI, plus Deep Research, Nano Banana Pro, Veo 3.1 access, 1,000 monthly AI credits, and 5 TB of Google One storage; includes YouTube Premium Lite; 50% off the first year for new subscribers."
2027
+ "notes": "Gemini 3.1 Pro and supporting multimodal features in the Gemini app, Antigravity quota refreshed every five hours up to a weekly limit, plus Deep Research, Nano Banana Pro, Veo 3.1 access, 1,000 monthly AI credits, and 5 TB of Google One storage; includes YouTube Premium Lite; 50% off the first year for new subscribers."
2009
2028
  },
2010
2029
  {
2011
2030
  "provider": "Google",
@@ -2013,8 +2032,8 @@
2013
2032
  "monthly_usd": 99.99,
2014
2033
  "annual_usd": null,
2015
2034
  "surface_funded": [
2016
- "gemini-app",
2017
- "gemini-cli"
2035
+ "antigravity",
2036
+ "gemini-app"
2018
2037
  ],
2019
2038
  "notes": "Developer-focused Ultra tier (May 2026): 5x Pro usage limits in Gemini app and Google Antigravity, Gemini 3.5 Flash integration, priority access to Antigravity, 20 TB cloud storage, and YouTube Premium individual plan."
2020
2039
  },
@@ -2024,8 +2043,8 @@
2024
2043
  "monthly_usd": 199.99,
2025
2044
  "annual_usd": null,
2026
2045
  "surface_funded": [
2027
- "gemini-app",
2028
- "gemini-cli"
2046
+ "antigravity",
2047
+ "gemini-app"
2029
2048
  ],
2030
2049
  "notes": "Highest access to Gemini 3.1 Pro, Deep Think, Project Genie, Veo 3.1 with 25,000 monthly AI credits, $100/month Google Cloud credits, and 30 TB Google One storage (price reduced from $249.99 to $199.99 at Google I/O 2026)."
2031
2050
  },
@@ -1037,7 +1037,7 @@
1037
1037
  headline-benchmarks="AA Intelligence Index 30.1 (adaptive reasoning, max effort); HLE 33.6%; SciCode 50.1; Terminal-Bench 4.0 3.0; AA-LCR 0.800; LMArena Text Elo 1457.7 (#41); AA-Omniscience 12.4"
1038
1038
  pricing-notes="Hidden by default; Requires Max Mode on legacy request-based plans; Up to 1M tokens with extended context at the same per-token rates (no long-context surcharge)"
1039
1039
  prose-generated="best-for headline-benchmarks"
1040
- best-for="Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models. Superseded by Sonnet 5." superseded-by="claude-sonnet-5" superseded-on="2026-09-24"/>
1040
+ best-for="Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models."/>
1041
1041
 
1042
1042
  <model id="gpt-5.4" name="GPT-5.4"
1043
1043
  input-price-per-1m="$2.50" output-price-per-1m="$15.00"
@@ -1601,13 +1601,13 @@
1601
1601
  exposes-orchestration="no"
1602
1602
  best-for="Programmatic / scripted Gemini use with a Google API key. Pay-per-token at Google's published rates. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) shared across the 3.x and 2.5 generations — per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash). Powers the roadmodel SaaS free-tier surfaces (/recommend on Gemini 2.5 Flash; /roadmap on Gemini 2.5 Flash with 3.1 Pro escalation)." />
1603
1603
  <method id="gemini-cli" name="Gemini CLI"
1604
- provider="google" billing="subscription-or-key"
1604
+ provider="google" billing="per-token"
1605
1605
  provider-jurisdiction="us"
1606
- requires="gemini-advanced-subscription OR google-api-key"
1606
+ requires="google-api-key"
1607
1607
  supports-models="gemini-3.1-pro,gemini-3-pro,gemini-3.8-flash,gemini-3.7-flash,gemini-3.6-flash,gemini-3.5-flash,gemini-3-flash,gemini-2.5-flash"
1608
1608
  exposes-max-mode="no" exposes-thinking="yes"
1609
1609
  exposes-orchestration="no"
1610
- best-for="Terminal-driven Gemini use; the CLI surface for multimodal and long-context Gemini work outside Cursor's pool. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) — per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash)." />
1610
+ best-for="Since 2026-06-18 the Gemini CLI serves only paid Gemini API keys and Gemini Code Assist Standard / Enterprise licences: Google AI Pro, Ultra and free personal accounts moved to Antigravity (`antigravity`). Terminal-driven Gemini use; the CLI surface for multimodal and long-context Gemini work outside Cursor's pool. Exposes the discrete Gemini thinking-level dial (`minimal`, `low`, `medium`, `high`) — per-model support varies (Gemini 3 Pro is low/high only; Gemini 3.1 Pro / 3.7 Flash / 3.8 Flash / 2.5 Pro / 2.5 Flash / 2.5 Flash-Lite are low/medium/high without `minimal`; `minimal` appears on Flash-line models such as 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite Image, and 3 Flash)." />
1611
1611
  <method id="gemini-app" name="Gemini (web / app)"
1612
1612
  provider="google" billing="subscription-included"
1613
1613
  provider-jurisdiction="us"
@@ -1616,6 +1616,14 @@
1616
1616
  exposes-max-mode="no" exposes-thinking="yes"
1617
1617
  exposes-orchestration="no"
1618
1618
  best-for="Chat-driven Gemini use under the Gemini Advanced subscription budget." />
1619
+ <method id="antigravity" name="Antigravity"
1620
+ provider="google" billing="subscription-included"
1621
+ provider-jurisdiction="us"
1622
+ requires="google-ai-pro-or-ultra-subscription"
1623
+ supports-models="gemini-3.1-pro,gemini-3.8-flash,gemini-3.7-flash,gemini-3.6-flash,sonnet-4.6,gpt-oss-120b"
1624
+ exposes-max-mode="no" exposes-thinking="yes"
1625
+ exposes-orchestration="no"
1626
+ best-for="Google's agent-first coding surface: the Antigravity desktop IDE, its VS Code panel and the `agy` CLI, which replaced the Gemini CLI for personal Google accounts on 2026-06-18. Google AI Pro and Ultra pay for it: quota refreshes every five hours up to a weekly limit (the free tier refreshes weekly), and both can buy AI credits for overage. It serves the Gemini models and, on Google's quota, Claude Sonnet 4.6 and Claude Opus 4.6 with thinking and gpt-oss-120b (antigravity.google/docs/models); Claude Opus 4.6 is not in this catalog. Gemini models take a Low / Medium / High reasoning level, set in the model id (`<model>-low` / `-medium` / `-high`) or with `agy --effort`; the Claude and GPT-OSS models carry their own thinking setting. It bills a Google AI plan or a Google Cloud project, not a Gemini API key." />
1619
1627
  <method id="xai-api" name="xAI API"
1620
1628
  provider="xai" billing="per-token"
1621
1629
  provider-jurisdiction="us"
@@ -207,10 +207,10 @@ and adjust the rebuild rules in [`update/prompt.md`](../update/prompt.md).
207
207
  | ChatGPT Pro ($100) | $100 | — | OpenAI | chatgpt-app, codex-cli | Same model suite as Pro $200 (GPT-5.6 Sol, GPT-5.5, o1 Pro mode) at 5x Plus usage limits; Codex access included with promotional 10x multiplier through May 31, 2026. |
208
208
  | ChatGPT Pro ($200) | $200 | — | OpenAI | chatgpt-app, codex-cli | Same core Pro features as Pro $100 (Pro models, Codex, deep research, image creation, memory, file uploads) with more usage than Pro $100; new subscriptions are open again, and new subscriptions not eligible for grandfathering get a lower usage allowance than the previous Pro 200. |
209
209
  | ChatGPT Pro ($500) | $500 | — | OpenAI | chatgpt-app, codex-cli | Highest-usage ChatGPT Pro tier (more included usage than Pro $100 and Pro $200); includes Astra Ultrafast plus the core Pro features (Pro models, Codex, deep research, image creation, memory, file uploads). |
210
- | Google AI Plus | $4.99 | — | Google | gemini-app, gemini-cli | Entry-paid Google AI tier with 2x higher usage limits than Free in the Gemini app, access to Gemini 3.1 Pro / Nano Banana Pro / Daily Brief / Gemini Omni video generation, 200 Google Flow Credits, and 400 GB of cloud storage (price cut from $7.99 to $4.99 on 2026-06-08; storage doubled from 200 GB to 400 GB). |
211
- | Google AI Pro | $19.99 | $199.99 | Google | gemini-app, gemini-cli | Gemini 3.1 Pro and supporting multimodal features in the Gemini app and Gemini CLI, plus Deep Research, Nano Banana Pro, Veo 3.1 access, 1,000 monthly AI credits, and 5 TB of Google One storage; includes YouTube Premium Lite; 50% off the first year for new subscribers. |
212
- | Google AI Ultra ($100) | $99.99 | — | Google | gemini-app, gemini-cli | Developer-focused Ultra tier (May 2026): 5x Pro usage limits in Gemini app and Google Antigravity, Gemini 3.5 Flash integration, priority access to Antigravity, 20 TB cloud storage, and YouTube Premium individual plan. |
213
- | Google AI Ultra ($200) | $199.99 | — | Google | gemini-app, gemini-cli | Highest access to Gemini 3.1 Pro, Deep Think, Project Genie, Veo 3.1 with 25,000 monthly AI credits, $100/month Google Cloud credits, and 30 TB Google One storage (price reduced from $249.99 to $199.99 at Google I/O 2026). |
210
+ | Google AI Plus | $4.99 | — | Google | gemini-app | Entry-paid Google AI tier with 2x higher usage limits than Free in the Gemini app, access to Gemini 3.1 Pro / Nano Banana Pro / Daily Brief / Gemini Omni video generation, 200 Google Flow Credits, and 400 GB of cloud storage (price cut from $7.99 to $4.99 on 2026-06-08; storage doubled from 200 GB to 400 GB). |
211
+ | Google AI Pro | $19.99 | $199.99 | Google | gemini-app, antigravity | Gemini 3.1 Pro and supporting multimodal features in the Gemini app, Antigravity quota refreshed every five hours up to a weekly limit, plus Deep Research, Nano Banana Pro, Veo 3.1 access, 1,000 monthly AI credits, and 5 TB of Google One storage; includes YouTube Premium Lite; 50% off the first year for new subscribers. |
212
+ | Google AI Ultra ($100) | $99.99 | — | Google | gemini-app, antigravity | Developer-focused Ultra tier (May 2026): 5x Pro usage limits in Gemini app and Google Antigravity, Gemini 3.5 Flash integration, priority access to Antigravity, 20 TB cloud storage, and YouTube Premium individual plan. |
213
+ | Google AI Ultra ($200) | $199.99 | — | Google | gemini-app, antigravity | Highest access to Gemini 3.1 Pro, Deep Think, Project Genie, Veo 3.1 with 25,000 monthly AI credits, $100/month Google Cloud credits, and 30 TB Google One storage (price reduced from $249.99 to $199.99 at Google I/O 2026). |
214
214
  | Cursor Pro | $20 | $192 | Cursor | cursor | Two monthly usage pools — Cursor Models (generous included usage for Grok 4.7, Grok 4.6, Grok 4.5, Composer 2.5) and Other Models (third-party models at API price); on-demand overage billed at API rates. |
215
215
  | Cursor Pro+ | $60 | $576 | Cursor | cursor | Same model coverage as Pro at roughly 3x the OpenAI / Claude / Gemini usage budget. |
216
216
  | Cursor Ultra | $200 | $1920 | Cursor | cursor | Same model coverage as Pro at roughly 20x the OpenAI / Claude / Gemini usage budget; priority access to new features. |
@@ -15,6 +15,13 @@ per-token prices) so that:
15
15
  The weekly catalog refresh in [`update/prompt.md`](../update/prompt.md)
16
16
  **must not** modify this file.
17
17
 
18
+ **Describe access, not models.** This file says what you pay for and how you
19
+ work. Leave model versions out: the catalog decides which models qualify and
20
+ keeps up with every release, so a version written here goes stale the day its
21
+ successor ships. Name a model family (Opus, Sonnet) only to state a
22
+ preference; the selector reads it as that family's newest model the catalog
23
+ does not mark superseded.
24
+
18
25
  ## How the selector consumes this file
19
26
 
20
27
  When [`model-selector.txt`](model-selector.txt) is referenced, the AI
@@ -186,8 +193,9 @@ dial the chosen platform lacks gets NO line at all — never `Off`, never
186
193
  - **Thinking** — default `On` wherever extended reasoning helps; `Off`
187
194
  only for purely mechanical steps where latency matters. It never
188
195
  carries an effort word or a number; `THINKING: Max` is not a setting
189
- any surface can apply. Claude Fable 5 cannot disable extended
190
- thinking, so never declare `Off` for it.
196
+ any surface can apply. Some models cannot disable extended
197
+ thinking (the Claude Fable line, for one), so never declare `Off`
198
+ for those.
191
199
  - **Codex / OpenAI** — the surface labels its dial "Intelligence" in its
192
200
  own UI, but the emitted field is still `EFFORT` (same value, different
193
201
  name). It exposes no separate thinking toggle and no Max Mode.
@@ -204,8 +212,8 @@ dial the chosen platform lacks gets NO line at all — never `Off`, never
204
212
  than a step inside one. By default each phase's **Phase roadmap
205
213
  settings** are picked quality first: the strongest planning model you
206
214
  can reach, at its deepest reasoning rung (`Max` in Claude Code). State
207
- a rule here to override that, for example "Phase roadmaps: Opus 5.5 at
208
- `XHigh`" to keep more of a weekly pool.
215
+ a rule here to override that, for example "Phase roadmaps: one rung below
216
+ the deepest" to keep more of a weekly pool.
209
217
 
210
218
  ## Budget priority and speed posture
211
219
 
@@ -232,14 +240,11 @@ the frontier tier on every posture.
232
240
  **Speed posture:** speed is NOT a valued dimension — workflows are
233
241
  batch / asynchronous, not latency-sensitive. Apply these rules:
234
242
 
235
- - **Never recommend a "Fast" variant** of a model (e.g., Opus 4.7
236
- Fast, GPT-5.4 Fast, GPT-5 Fast, Claude 4.6 Opus Fast mode) when a
237
- standard non-fast variant of the same base model is available.
238
- Fast variants in Cursor's catalog typically charge 2x output for
239
- marginal speed gain (e.g., "Fast mode is 15% faster with 2x
240
- pricing", "Claude 4.6 Opus Fast mode" at $30/$150 input/output vs.
241
- standard Opus at $5/$25) — the speed-for-cost trade is never worth
242
- it here.
243
+ - **Never recommend a "Fast" variant** of a model (e.g., an Opus Fast
244
+ or a GPT Fast mode) when a standard non-fast variant of the same base
245
+ model is available. Fast variants typically charge 2x or more on
246
+ output for a marginal speed gain — the speed-for-cost trade is never
247
+ worth it here.
243
248
  - **In `<selection-algorithm>` Step 5** (cost tie-breaker), when the
244
249
  tied set contains both a standard and a Fast variant of the same
245
250
  base model, the standard variant always wins regardless of any
@@ -358,7 +363,7 @@ platforms.excluded: (none declared)
358
363
  Values are **access-method ids** from the `<access-methods>` block in
359
364
  `model-selector.txt` — e.g. `claude-code`, `cursor`, `codex-cli`,
360
365
  `chatgpt-app`, `claude-web`, `anthropic-api`, `openai-api`,
361
- `google-api`, `gemini-cli`, `deepseek-api`, `mistral-api` — not
366
+ `google-api`, `gemini-cli`, `antigravity`, `deepseek-api`, `mistral-api` — not
362
367
  display names ("Claude Code") and not provider names ("Anthropic").
363
368
  An operator who works only in the terminal might declare:
364
369
 
@@ -1,2 +1,2 @@
1
1
  # src/roadmodel/__init__.py
2
- __version__ = "0.2.54"
2
+ __version__ = "0.2.55"
@@ -391,6 +391,11 @@ _AGGREGATOR_PROVIDERS: frozenset[str] = frozenset({"cursor", "openrouter", "olla
391
391
  # pre-4.10 catalog gave (`cursor`) and the same-family backup guard keeps
392
392
  # catching Kimi-vs-Kimi. A per-model `maker` attribute would be the real fix.
393
393
  _AGGREGATOR_FALLBACK_ORDER: tuple[str, ...] = ("cursor", "openrouter", "ollama")
394
+ # Access methods that belong to a real maker but also host OTHER makers' models
395
+ # never decide a maker: Antigravity is Google's, and it serves Claude Sonnet 4.6
396
+ # and gpt-oss-120b too. A Gemini model's maker still resolves through Google's
397
+ # other methods.
398
+ _HOST_METHODS: frozenset[str] = frozenset({"antigravity"})
394
399
 
395
400
  # Providers whose access method is a LOCAL runtime (billing `local`). A model
396
401
  # whose only reachable platform is one of these is never a usable backup
@@ -425,6 +430,8 @@ def model_provider(model_ref: str) -> str | None:
425
430
  for method in methods:
426
431
  if not isinstance(method, dict):
427
432
  continue
433
+ if method.get("id") in _HOST_METHODS:
434
+ continue
428
435
  if model_id in method.get("supports_models", []):
429
436
  provider = method.get("provider")
430
437
  if isinstance(provider, str) and provider:
@@ -620,6 +627,10 @@ def _tiers_funding_surface(subscription_tiers: list[Any], surface_id: str) -> li
620
627
  return tiers
621
628
 
622
629
 
630
+ # A user-context price matches a catalog tier's within this many dollars.
631
+ _PRICE_MATCH_USD = 1.0
632
+
633
+
623
634
  def _match_active_tier(
624
635
  catalog_tiers: list[dict[str, Any]],
625
636
  active_subscriptions: list[tuple[str, str, float | None]],
@@ -633,10 +644,12 @@ def _match_active_tier(
633
644
  continue
634
645
  if sub_provider and sub_provider != catalog_provider:
635
646
  continue
647
+ # The price tells same-name tiers apart (Ultra $100 / $200), and
648
+ # a hand-written row rounds it: "$20" is the $19.99 Google AI Pro.
636
649
  if (
637
650
  isinstance(catalog_monthly, (int, float))
638
651
  and sub_monthly is not None
639
- and float(catalog_monthly) != sub_monthly
652
+ and abs(float(catalog_monthly) - sub_monthly) >= _PRICE_MATCH_USD
640
653
  ):
641
654
  continue
642
655
  return catalog_tier
@@ -35,8 +35,9 @@ Score, per candidate (model reached through a specific access method)::
35
35
  stakes (a hard or novel task weighs cost less, because a failed attempt
36
36
  costs more than the price gap).
37
37
 
38
- Hard filters run first: availability, allowed jurisdictions, the operator's
39
- platform allow / deny lists. Funding is read from the user-context exactly as
38
+ Hard filters run first: availability, supersession (a superseded model
39
+ leaves while its successor is available, as it leaves the recommender's),
40
+ allowed jurisdictions, the operator's platform allow / deny lists. Funding is read from the user-context exactly as
40
41
  :mod:`roadmodel.cost` reads it (Active subscriptions / Active API keys / Local
41
42
  models) plus the ``Consumption headroom`` line and the ``Usage-pool status``
42
43
  table; an unfunded platform is never chosen while a funded one exists.
@@ -175,7 +176,7 @@ DEFAULT_K: Final[float] = 30.0
175
176
  # Coding-agent surfaces get a nudge on coding / agentic / planning work (that
176
177
  # is where roadmap steps run) when costs tie; chat apps and raw APIs do not.
177
178
  AGENT_SURFACES: Final[frozenset[str]] = frozenset(
178
- {"claude-code", "codex-cli", "cursor", "gemini-cli"}
179
+ {"claude-code", "codex-cli", "cursor", "gemini-cli", "antigravity"}
179
180
  )
180
181
  AGENT_CATEGORIES: Final[frozenset[str]] = frozenset({"coding", "agentic", "planning"})
181
182
  # Platform tie-break when scores are equal: a paid-for subscription surface,
@@ -781,6 +782,7 @@ def rank(
781
782
  methods = [m for m in cat.get("access_methods", []) if isinstance(m, dict)]
782
783
  method_ids = {str(m.get("id", "")) for m in methods}
783
784
  families = frozenset(_family_words(cat))
785
+ catalog_ids = {str(m.get("id", "")) for m in cat.get("models", []) if isinstance(m, dict)}
784
786
  candidates: list[Candidate] = []
785
787
  excluded: list[dict[str, str]] = []
786
788
  # A platform list is only as good as its ids: unknown tokens are reported
@@ -803,6 +805,12 @@ def rank(
803
805
  if model_id in unavailable:
804
806
  excluded.append({"model": model_id, "reason": "unavailable"})
805
807
  continue
808
+ # The successor is the same maker's, no dearer, and rated at least as
809
+ # high everywhere; a benched successor keeps the older model in play.
810
+ successor = str(model.get("superseded_by") or "")
811
+ if successor and successor in catalog_ids and successor not in unavailable:
812
+ excluded.append({"model": model_id, "reason": f"superseded by {successor}"})
813
+ continue
806
814
  if str(model.get("jurisdiction", "")).lower() not in juris:
807
815
  excluded.append({"model": model_id, "reason": "jurisdiction"})
808
816
  continue
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes