Add Bing grounding tool pricing for Azure AI Foundry (1.4¢/call)

[akhilmmenon]

Summary

Adds Bing grounding tool pricing for Azure AI Foundry (azure-ai) models that support web-grounded tools, so these native tool calls are billed correctly.

• bing_grounding: price: 1.4 (1.4¢/call)
• bing_custom_search: price: 1.4 (1.4¢/call)

Added to all 68 additional_units blocks that already price web_search (i.e. the web-capable Foundry models that can emit Bing tool calls). Pure additive change (+408 / -0 lines), no reformatting.

Pricing basis: Grounding with Bing Search & Custom Search are both $14 / 1,000 transactions = $0.014/call = 1.4¢, per Grounding with Bing pricing.

Companion gateway change (counting logic for all native tools): Portkey-AI/gateway-enterprise-node#2233
Asana: https://app.asana.com/1/11915891072957/project/1215728220879780/task/1218810414553424

Why only Bing (and not the other 17 native tools)?

The gateway counts all 19 native platform tools and writes them to tokens_raw, but billing only activates for keys that have a price here. We added prices only for tools with a clean, flat, per-call list price that fits this repo's price × call-count model. The rest are session-, token-, or credit-based, or have no published per-call price.

Tool (unit key)	Pricing basis in vendor docs	Per-call?	Decision	Reference
web_search	$10 / 1k calls	✅	Already priced (1¢)	OpenAI pricing
file_search	$2.50 / 1k calls	✅	Already priced (0.25¢)	OpenAI pricing
bing_grounding	$14 / 1k transactions	✅	Priced here (1.4¢)	Grounding with Bing
bing_custom_search	$14 / 1k transactions	✅	Priced here (1.4¢)	Grounding with Bing
mcp	Model tokens only — no per-call fee	✅ ($0)	Not added (already $0)	OpenAI MCP guide
computer_use	Model tokens only — no hosted-tool fee	✅ ($0)	Not added (already $0)	OpenAI pricing · eesel 2026
code_interpreter	$0.03–$1.92 per 20-min container session (by size)	❌ session	Not added — not per-call	OpenAI pricing
shell	Same container-session pricing as code_interpreter	❌ session	Not added — not per-call	OpenAI pricing
local_shell	Same container-session pricing as code_interpreter	❌ session	Not added — not per-call	OpenAI pricing
image_generation	Token-based (e.g. $30/1M image-output tokens)	❌ tokens	Not added — token-based	gpt-image-2.5
azure_ai_search (Foundry IQ)	Token-based agentic retrieval (per 1M tokens; 50M free/mo)	❌ tokens	Not added — token-based	Foundry IQ pricing
sharepoint_grounding	Credit/license-based (~12 credits/query)	❌ credits	Not added — no flat per-call	Foundry Agent pricing · M365 cost tool
fabric_dataagent	Separate charges/licenses, credit-based	❌ credits	Not added — no flat per-call	Foundry Agent pricing
apply_patch	No standalone per-call price published	❌ none	Not added — no price	OpenAI pricing
tool_search	No standalone per-call price published	❌ none	Not added — no price	OpenAI pricing
browser_automation	No standalone per-call price published (preview)	❌ none	Not added — no price	Foundry Agent pricing
memory_search	No standalone per-call price published	❌ none	Not added — no price	Foundry Agent pricing
memory_command	No standalone per-call price published (preview)	❌ none	Not added — no price	Foundry Agent pricing
a2a	No standalone per-call price (billed via invoked agent)	❌ none	Not added — no price	Foundry Agent pricing

Note: list prices as of research date (Oct 2026). The *_preview tools (bing_custom, sharepoint, fabric, browser_automation, memory_command, a2a) are preview features whose pricing may still change.

[Copilot]

Copilot review overview

🟢 Approval recommended

The JSON is valid, and all 68 web-search pricing blocks consistently include both Bing tools at the documented rate.

Review effort: Balanced
Findings: None

What changed in this PR

Adds Bing grounding tool pricing to all Azure AI Foundry configurations that already support web search billing.

Changes:

• Prices bing_grounding and bing_custom_search at 1.4¢ per call.
• Updates all 68 applicable pricing blocks.

File	Description
pricing/​azure-ai.json	Adds Bing tool pricing to web-capable models.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.