Summary

Daily model inventory check for 2026-07-01. Three new models were found requiring updates to models.json and model_aliases.json.

• Providers queried: OpenAI (live API), Anthropic (live API), Gemini (live API), Copilot (api-proxy reflect)
• Total models found: OpenAI 217, Anthropic 9, Gemini 50, Copilot 32
• Proposed alias changes: 3 new aliases added (fable, gpt-5.6, gemini-omni); large alias updated
• Pricing gaps found: 2 new models added to models.json (claude-sonnet-5, claude-fable-5)
• Changes committed: 9f57af1

Provider Model Counts

Raw API Fields Discovered

OpenAI

Fields available: id, object, created, owned_by. Minimal metadata — no pricing, context limits, or capabilities exposed directly. Use owned_by and model-ID naming conventions for tier inference.

Anthropic

Rich fields: id, display_name, type, created_at, max_input_tokens, max_tokens, capabilities (structured object with batch, citations, code_execution, context_management, effort, image_input, pdf_input, structured_outputs, thinking). Capability flags (effort, thinking) are useful for reasoning/extended-thinking detection.

Gemini

Fields: name, version, displayName, description, inputTokenLimit, outputTokenLimit, supportedGenerationMethods (array), temperature, topP, topK, maxTemperature, thinking (boolean). Token limits are useful for pricing-tier inference. thinking: true indicates reasoning-capable models.

Copilot (reflect endpoint)

The reflect endpoint returns a flat list of model ID strings (not objects). No billing multiplier or capability metadata available from this endpoint. The model_api_mapping section in reflect provides API endpoint routing (responses vs chat_completions) for OpenAI and Anthropic families — this is already reflected in wire_api fields in models.json.

models.json Pricing Analysis

Missing from models.json (now added in commit 9f57af1)

Historical entries not currently returned

The following model IDs appear in models.json but are absent from all live provider inventories. These are retained as historical records per policy:

claude-3-5-haiku-20241022, claude-3-5-haiku-latest, claude-3-5-sonnet-20240620, claude-3-5-sonnet-20241022, claude-3-7-sonnet-20250219, claude-3-haiku-20240307, claude-3-opus-20240229, claude-3-sonnet-20240229, claude-sonnet-4-0, claude-sonnet-4-20250514, claude-opus-4-0, claude-opus-4-20250514, gpt-5-search-api, o1-mini, o1-preview, claude-sonnet-4 (Copilot), raptor-mini (Copilot).

Inferred vs stored pricing discrepancies

No discrepancies found for existing entries. The Copilot reflect endpoint does not expose billing multipliers (returns string IDs only); the docs pricing table returned empty models array for this run.

Proposed Alias Updates

1. New fable alias

What: New fable alias for the Claude Fable 5 family.
Why: claude-fable-5 is a brand-new Anthropic model family (released 2026-06-09) that is not matched by any existing alias. The existing sonnet, opus, and haiku patterns are name-specific and do not catch fable. Available on both Anthropic direct and Copilot (per models.dev).
Syntax:

"fable": [
  "copilot/*fable*",
  "anthropic/*fable*"
]

2. New gpt-5.6 alias

What: Version-specific alias for the gpt-5.6 generation.
Why: Live OpenAI API returns gpt-5.6, gpt-5.6-luna, gpt-5.6-sol, gpt-5.6-terra. While technically covered by the broad gpt-5 wildcard (openai/gpt-5*), a version-specific alias is consistent with the existing pattern for gpt-5.1–gpt-5.5 and allows targeted pinning.
Syntax:

"gpt-5.6": [
  "copilot/gpt-5.6*",
  "openai/gpt-5.6*"
]

3. New gemini-omni alias

What: Explicit alias for Gemini Omni Flash Preview.
Why: gemini-omni-flash-preview is a new Gemini model family (distinct from standard Flash). While the gemini-flash wildcard (gemini-*flash*) would technically match it, the omni family is architecturally distinct (multimodal/audio focus, 131K context vs 1M for standard Flash). An explicit alias avoids ambiguity.
Syntax:

"gemini-omni": [
  "copilot/gemini-omni*",
  "google/gemini-omni*",
  "gemini/gemini-omni*"
]

4. Updated large alias

What: large now lists fable first as the flagship option.
Why: Claude Fable 5 is Anthropicâ€TMs new highest-capability model ($10/$50 per MTok, 1M context, extended reasoning). It should resolve before sonnet for large-tier requests.
Syntax:

"large": [
  "fable",
  "sonnet",
  "gpt-5-pro",
  "gpt-5",
  "gemini-pro"
]

Notes

• gpt-4o, gpt-4o-mini, gpt-4.1, and gpt-5.4-nano remain in the Copilot live model list (reflect) but are intentionally excluded from models.json updates per task policy.
• claude-fable-5 is not yet live in the Copilot reflect endpoint (not in the 32-model list); it was added to github-copilot provider in models.json based on models.dev catalog data. Should be verified when Copilot onboards it.
• gpt-5.6 variants (gpt-5.6-luna, -sol, -terra) are in the live OpenAI API but absent from models.dev — likely experimental. The gpt-5.6 alias covers them via wildcard.
• Historical entries in models.json (deprecated model IDs) are retained per policy; no removals proposed.
• gemini-omni-flash-preview has 131K context and thinking: true — it is functionally a reasoning-capable flash model, not just a standard Flash variant, hence the dedicated alias.

network:
  allowed:
    - defaults
    - "models.dev"
    - "proxy.golang.org"

Generated by 📦 Daily Model Inventory Checker · 140.9 AIC · ⌖ 11 AIC · ⊞ 7.7K · ◷

• expires on Jul 7, 2026, 8:00 PM UTC-08:00