[model-inventory] Model alias inventory update - 2026-08-16

[github-actions]

Summary

Live provider inventories are ahead of the cached models.json payload for a handful of Copilot-served models, and one new Gemini Flash generation lacks a dedicated alias for consistency with existing gemini-3.5-flash / gemini-3.6-flash entries. pkg/workflow/data/model_aliases.json wildcard patterns already cover all newly observed models via family globs (*sonnet*, *opus*, *grok*, gemini-*flash*), so no alias is broken today, but the pricing table (pkg/cli/data/models.json / mirrored actions/setup/js/models.json) is missing 3 models seen live.

• Providers queried: OpenAI, Anthropic, Gemini, Copilot SDK, Copilot (reflect)
• Total models found: 223 (openai) + 9 (anthropic) + 50 (gemini) + 24 (copilot-sdk) + 36 (copilot reflect)
• Proposed alias changes: 1 (new gemini-3.7-flash alias for consistency)
• Pricing gaps found: 3 missing from models.json (grok-4.6, gemini-3.7-flash, auto)

Provider Model Counts

Provider	Models Available	Status
openai (API)	223	✅ ok
anthropic (API)	9	✅ ok
gemini (API)	50	✅ ok
copilot-sdk	24	✅ ok
copilot (reflect, /reflect proxy)	36	✅ ok
models.dev (github-copilot)	cross-check only	✅ ok

Raw API Fields Discovered

• Copilot reflect (/reflect): model_metadata[].pricing.default.{input,output,cachedInput,cacheWrite} and .long_context.* with a threshold field for long-context tier switch. Also api_version, observed_at.
• Copilot SDK: capabilities.limits.{max_context_window_tokens,max_output_tokens,max_prompt_tokens}, capabilities.supports.{vision,tool_calls,streaming,structured_outputs,reasoning_effort}, billing.tokenPrices.{inputPrice,outputPrice,cachePrice,cacheReadPrice,cacheWritePrice,contextMax,maxPromptTokens,longContext.*}, modelPickerCategory, modelPickerPriceCategory, policy.state/policy.terms (rollout status text).
• Anthropic API: only 9 models returned (all claude-*-4/4.5/5 generations) — no legacy IDs, confirming the currently-stored legacy Claude 3.x IDs in models.json are historical only.
• Gemini API: 50 models incl. many preview/experimental lines (antigravity-preview, deep-research-*, nano-banana-pro-preview, veo-3.1-*, lyria-3-*) not modeled as Copilot-servable chat models — out of scope for models.json (image/video/audio generation models).
• models.dev (github-copilot): normalized IDs used as the make refresh-models-json refresh source; confirms grok-4.6 and gemini-3.7-flash are present upstream and ready to sync.
• Copilot docs pricing table: exposes Category (Lightweight/Versatile/Powerful) and Tier/Threshold (Default vs Long context split) alongside $ input/output/cached-input/cache-write prices — all GPT-5.x rows cross-checked exactly against stored cost values in models.json (no discrepancies).

models.json Pricing Analysis

Missing from models.json

Model ID	Provider	Inferred Pricing	Basis
grok-4.6	copilot	input $2.00 / output $6.00 per 1M (scaled from SDK tokenPrices using same $/unit ratio as stored grok-4.5)	Copilot SDK billing.tokenPrices (identical to grok-4.5: inputPrice 200, outputPrice 600)
gemini-3.7-flash	copilot	input $0.75 / output $3.75 per 1M	Copilot SDK billing.tokenPrices (inputPrice 75, outputPrice 375 — identical to stored gemini-3.6-flash pricing)
auto	copilot	not a priced model — router alias with a 10% discount modifier	Copilot SDK billing.discountPercent: 10; already referenced literally in model_aliases.json ("auto": ["copilot/auto", "large"]) but has no models.json cost entry

Note: make refresh-models-json (models.dev-sourced) already picks up grok-4.6 and gemini-3.7-flash automatically, plus several other new entries (e.g. gpt-5.2-pro, gpt-5.3-codex-spark, gpt-image-1.5/-2, gpt-realtime-2.1) not required by this run's live inventory but consistent with it. Running the refresh is the recommended path rather than hand-editing.

Historical entries not currently returned

These remain in models.json but were absent from all live inventories this run (kept as historical records per policy, not proposed for removal):

claude-fable-5, claude-opus-4.5, claude-opus-4.6, claude-sonnet-4, gemini-2.5-pro, gemini-3-flash-preview, gemini-3.1-pro-preview, gemini-3.5-flash, gemini-3.6-flash, gpt-5.2, gpt-5.2-codex, gpt-5.4-nano (intentionally deprecated, excluded from gap analysis per policy), kimi-k2.7-code, kimi-k3, raptor-mini.

Also note: legacy Claude 3.x IDs (claude-3-5-sonnet-*, claude-3-opus-*, etc.) and gpt-4-turbo/o1-mini/o1-preview remain in models.json as historical entries; the Anthropic and OpenAI APIs no longer list them, consistent with prior runs.

Inferred vs stored pricing discrepancies

No pricing discrepancies found. All GPT-5.x family entries (gpt-5-mini, gpt-5.3-codex, gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.5, gpt-5.6-luna, gpt-5.6-sol, gpt-5.6-terra) matched exactly against the live GitHub Docs pricing table (default and long-context tiers). grok-4.5 and gemini-3.6-flash stored costs matched their Copilot SDK tokenPrices values.

Proposed Alias Updates

What: Add a dedicated gemini-3.7-flash alias, mirroring the existing gemini-3.5-flash / gemini-3.6-flash per-generation aliases.
Why: gemini-3.7-flash is now live in the Copilot SDK inventory and models.dev catalog. It is already reachable through the generic gemini-flash wildcard alias, but every other recent Flash generation (3.1, 3.5, 3.6) has its own dedicated alias for pinning to an exact generation — 3.7 is the only recent gap.
Syntax:

{
  "gemini-3.7-flash": ["copilot/gemini-3.7*flash*", "google/gemini-3.7*flash*", "gemini/gemini-3.7*flash*"]
}

No other alias changes are proposed: grok-4.6 is already covered by the existing "grok": ["copilot/*grok*", "openai/*grok*"] wildcard, and all new Claude/GPT generations observed this run (claude-opus-5, claude-sonnet-5, gpt-5.6-*) are already matched by existing family-wildcard aliases (opus, sonnet, gpt-5.6).

Full Model Lists by Provider

Anthropic (9): claude-haiku-4-5-20251001, claude-opus-4-5-20251101, claude-opus-4-6, claude-opus-4-7, claude-opus-4-8, claude-opus-5, claude-sonnet-4-5-20250929, claude-sonnet-4-6, claude-sonnet-5

Copilot (reflect, 36): claude-haiku-4.5, claude-opus-4.7, claude-opus-4.8, claude-opus-5, claude-sonnet-4.5, claude-sonnet-4.6, claude-sonnet-5, gpt-3.5-turbo, gpt-3.5-turbo-0613, gpt-4, gpt-4-0613, gpt-4-o-preview, gpt-4.1, gpt-4.1-2025-04-14, gpt-41-copilot, gpt-4o, gpt-4o-2024-05-13, gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini, gpt-4o-mini-2024-07-18, gpt-5-mini, gpt-5.3-codex, gpt-5.4, gpt-5.4-mini, gpt-5.5, gpt-5.6-luna, gpt-5.6-sol, gpt-5.6-terra, grok-4.5, grok-4.6, mai-code-1-flash-picker, mai-code-1.1-flash, text-embedding-3-small, text-embedding-3-small-inference, text-embedding-ada-002, trajectory-compaction

Copilot SDK (24): auto, claude-sonnet-5, claude-opus-5, claude-opus-4.8, claude-opus-4.7, claude-sonnet-4.6, claude-sonnet-4.5, claude-haiku-4.5, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5.3-codex, gpt-5-mini, mai-code-1-flash-picker, gemini-3.7-flash, gemini-3.6-flash, gemini-3.5-flash, gemini-3.1-pro-preview, grok-4.5, grok-4.6, mai-code-1.1-flash

Gemini (50): full generative/embedding/image/video/audio inventory incl. gemini-2.5-*, gemini-3-*, gemini-3.1-*, gemini-3.5-flash, gemini-3.6-flash, gemini-3.7-flash, gemma-4-*, imagen-4.0-*, veo-3.1-*, lyria-3-*, nano-banana-pro-preview, deep-research-*, antigravity-preview-05-2026 — see /tmp/gh-aw/agent/model-inventory/artifacts/gemini-models/models.json for the complete list.

OpenAI (223): full catalog spans gpt-3.5-* through gpt-5.x/o1/o3/o4 reasoning families and embeddings — see /tmp/gh-aw/agent/model-inventory/artifacts/openai-models/raw.json for the complete list.

Notes

• pkg/cli/data/models.json and actions/setup/js/models.json remain identical (byte-for-byte) as required.
• Running make refresh-models-json produces a much larger diff than the 3 gaps identified above because the upstream models.dev catalog also refreshes many unrelated provider entries (new OpenAI reasoning/image models, etc.) beyond what this run's live-API inventory directly confirms. Recommend running the refresh as routine maintenance rather than a targeted fix tied to this issue, then spot-checking the 3 gap models above.
• gpt-4o-mini, gpt-4.1, gpt-4o, and gpt-5.4-nano continue to appear in Copilot reflect/docs data but are treated as intentionally deprecated per standing policy — excluded from gap tables and not proposed for restoration.
• Copilot billing table (Step 2.5) only lists the GPT-5.x family (no Claude/Gemini/Grok rows) — Claude/Gemini/Grok pricing validation relied on Copilot SDK + reflect data only.
• No stale alias patterns were found; all existing globs still match at least one live model.

Warning

Firewall blocked 7 domains

The following domains were blocked by the firewall during workflow execution:

• accounts.google.com
• android.clients.google.com
• clients2.google.com
• collector.githubapp.com
• contentautofill.googleapis.com
• www.google.com
• www.gstatic.com

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "accounts.google.com"
    - "android.clients.google.com"
    - "clients2.google.com"
    - "collector.githubapp.com"
    - "contentautofill.googleapis.com"
    - "www.google.com"
    - "www.gstatic.com"

See Network Configuration for more information.

Generated by 📦 Daily Model Inventory Checker · auto · 73.1 AIC · ⌖ 4.93 AIC · ⊞ 13K · ◷

• expires on Aug 22, 2026, 4:04 PM UTC-08:00