azure-models-for-devs

Azure models available for Brainforge devs

Models we deploy and how to use them. Last updated: 2026-04.

Model retirements and policy

• Official schedule: Azure OpenAI in Microsoft Foundry – Model retirements (text, audio, image, embedding).
• Policy: Prefer gpt-5 or above everywhere. Set AZURE_GPT_5_1_BASE_URL (and deploy gpt-5.1) to drive default chat, ticket pipeline, and CopilotKit to gpt-5.1.
• Key dates (text): gpt-4o standard retires 2026-03-31; other types 2026-10-01. gpt-4.1 / o4-mini 2026-10-14 / 2026-10-16. Replacements: gpt-4o → gpt-5.1, gpt-4.1 → gpt-5.
• Voice: gpt-4o-realtime-preview retires 2026-03-24; use gpt-realtime or gpt-realtime-mini.

Policy: East US 2 only

From here on, East US 2 (brainforge-openai-eastus2) is the only place we deploy Azure OpenAI models. brainforge-openai (East US) is legacy and we are deprecating it — no new deployments there; migrate anything still on East US onto East US 2 and point all env vars and code at East US 2. Use Deploy to East US 2 for models that used to exist only on East US.

Why East US 2: Official Foundry models sold directly by Azure (Global Standard) shows East US 2 has the broadest catalog availability for new standard deployments (e.g. gpt-5.3-codex, full Codex lineup). Legacy brainforge-openai (East US) is still deprecating and must not receive new deployments, but it can retain existing chat deployments—including gpt-5.4 and gpt-5.4-mini (verified az cognitiveservices account deployment list 2026-04). Prefer East US 2 for all new work; use East US only for migration gaps or when tooling (e.g. Foundry content controls) forces that resource. OpenCode CLI + Desktop use East US only (azure-eastus/*); see opencode-cli-brainforge.md.

Resources

Resource	Region	Status
brainforge-openai-eastus2	East US 2	Use this only. All platform, Slack, Codex, GitHub Actions.
brainforge-eastus2-resource	East US 2	AIServices (realtime + model-catalog OSS, e.g. DeepSeek, Llama). Different endpoint and API key than brainforge-openai-eastus2.
brainforge-openai	East US	Legacy — deprecating. No new deployments. Exists only until migration off East US is complete.

Deployments

Verified via Azure CLI (East US 2: 2026-03; East US legacy: 2026-04). Run: az cognitiveservices account deployment list -g brainforge -n <resource-name> -o table.

brainforge-openai-eastus2 (East US 2) — use only this

Deployment	Type
gpt-5.2, gpt-5.1, gpt-4o, gpt-5-mini, gpt-5.4, gpt-5.4-mini	Chat
gpt-5.2-codex	Codex / Responses API
gpt-5.3-codex, gpt-5-mini-codex	Codex / Responses API
text-embedding-3-large	Embeddings

brainforge-eastus2-resource (East US 2) — AIServices / catalog OSS

Deployment	Type
deepseek-v3-2	Chat (DeepSeek V3.2)
llama-3-3-70b-instruct	Chat (Llama 3.3 70B Instruct)
gpt-4o-realtime-preview, gpt-5.2-chat, gpt-realtime-mini	Realtime / voice (existing)

Deploy to East US 2

To match the old region and stop using East US, deploy the following on brainforge-openai-eastus2 (Azure portal / AI Model Catalog) if you need them. After deployment, add them to the table above.

Model / deployment	Type	Notes
o4-mini	Chat	Was on East US only.
gpt-4.1	Chat	Was on East US only.
gpt-4o-mini	Chat	Was on East US only.
Kimi-K2.5	Chat	Was on East US only.
DeepSeek-V3-0324	Chat	Was on East US only.
text-embedding-ada-002	Embeddings	Was on East US only.
gpt-4o-zoom-backfill	(internal)	Deploy only if still used.
gpt-realtime-mini (or gpt-realtime)	Realtime/voice	For voice/case-study; deploy if catalog has it in East US 2.
gpt-5.3-codex	Codex	When available; deploy for Codex.

brainforge-openai (East US) — deprecated

Do not add new deployments or new production usage. Listed for migration audit and edge cases (e.g. when East US 2 / Foundry workflows still point here temporarily).

Deployment	Type
gpt-4o, gpt-4.1, o4-mini, gpt-5-mini, gpt-5.1, gpt-4o-mini	Chat
gpt-5.4, gpt-5.4-mini	Chat (present on this resource as of 2026-04 CLI list; still migrate to East US 2 for new apps)
Kimi-K2.5	Chat (Kimi)
DeepSeek-V3-0324	Chat
text-embedding-3-large, text-embedding-ada-002	Embeddings
gpt-4o-zoom-backfill	(internal/backfill)
gpt-4o-realtime-preview, gpt-realtime-mini	Realtime/voice

OpenCode (CLI + Desktop): repo-root opencode.jsonc is MCP-only (no embedded provider / default model). Use azure-eastus/* + AZURE_OPENAI_EASTUS_API_KEY in ~/.config/opencode/opencode.json — start from opencode-user-config.azure-eastus-legacy.example.jsonc, opencode-user-config.example.json, and opencode-cli-brainforge.md.

Env vars (East US 2 only)

Use one key and one base URL for all Azure OpenAI usage: East US 2.

Var	Used by
BRAINFORGE_OPENAI_EASTUS2_API_KEY	Platform, Slack assistant, CopilotKit, GitHub Actions (auto-label, Codex). Prefer this; can be used as AZURE_OPENAI_API_KEY where a single key is required.
AZURE_OPENAI_EASTUS2_BASE_URL	East US 2 base URL (e.g. https://brainforge-openai-eastus2.openai.azure.com). Set all platform *_BASE_URL vars to this when using East US 2 only.
AZURE_OPENAI_API_KEY	Set to East US 2 key (same as above) for platform/Mastra so no code points at the old region.
AZURE_GPT_4o_REALTIME_*	Case-study voice; point at East US 2 once realtime is deployed there.
AZURE_AI_SERVICES_EASTUS2_KEY	API key for brainforge-eastus2-resource (az cognitiveservices account keys list -g brainforge -n brainforge-eastus2-resource). Catalog OSS / realtime on that resource (not the same key as brainforge-openai-eastus2).

Legacy East US (AZURE_OPENAI_EASTUS_API_KEY — OpenCode + narrow exceptions)

Var	Used by
AZURE_OPENAI_EASTUS_API_KEY	Key for brainforge-openai (East US). Use with OpenCode provider azure-eastus in repo opencode.jsonc. Do not substitute the East US 2 key.

Upgrades

Voice (case-study): gpt-4o-realtime-preview → gpt-realtime-mini

1. Deploy gpt-realtime-mini (or gpt-realtime) in brainforge-openai-eastus2 (East US 2) via Azure portal / AI Model Catalog.
2. Set AZURE_GPT_4o_REALTIME_BASE_URL to East US 2 base URL and AZURE_OPENAI_REALTIME_API_KEY to East US 2 key (or same as BRAINFORGE_OPENAI_EASTUS2_API_KEY).
3. Set AZURE_GPT_4o_REALTIME_MODEL_NAME=gpt-realtime-mini in platform env (token + trace routes use this).

GitHub Actions

• Auto-label: Prefer BRAINFORGE_OPENAI_EASTUS2_API_KEY (uses gpt-5.2). Fallback: OPENAI_API_KEY.
• Codex auto-fix: Set BRAINFORGE_OPENAI_EASTUS2_API_KEY; workflow uses gpt-5-mini-codex on East US 2.

Testing gpt-5.1

1. Env check (local or after pull-env)
   From apps/platform:

   SKIP_ENV_VALIDATION=1 npx tsx scripts/test-azure-model-config.ts

   Expect: Primary chat (CopilotKit): gpt-5.1 and gpt-5.1 is active when AZURE_GPT_5_1_BASE_URL is set.

2. Live app – CopilotKit

   • Start dev: npm run dev (or use deployed app).
   • Open dashboard, open Brainforge GPT / Meeting Assistant, send one short message.
   • In Azure Portal → brainforge-openai-eastus2 → Monitor → Logs or Metrics, filter by deployment gpt-5.1 and confirm a request around that time.

3. Ticket pipeline

   • Trigger the meeting-created webhook (e.g. node scripts/trigger-webhook-by-topic.mjs with a meeting that runs the pipeline) or create a meeting that generates tickets.
   • Again check Azure Monitor for gpt-5.1 (or gpt-5-mini / o4-mini if gpt-5.1 is not set) traffic.

4. List deployments

   ./scripts/azure-deployment-create.sh

   Confirm gpt-5.1 appears in the table.