Enterprise deep research APIs in 2026

Retention and data routes are key settings, on every plan

Zero retention does not wait for a contract. Set zero_retention on a key and a search made with it keeps no request text and no answer after the reply: it is not in the history, its answer is not reused, and its log lines carry neither. eu_egress_only sends every search from our server in Madrid and never through a proxy in another country. engine_rule keeps the key's queries off every search engine run from China, and a search key cannot change its own rule.

curl -X POST https://api.specstream.ai/v1/account/keys \
  -H "Authorization: Bearer <OAuth token>" \
  -H "Content-Type: application/json" \
  -d '{"name": "research-prod",
       "scopes": ["account:read", "finds:spend"],
       "spend_cap_cents": 500000,
       "zero_retention": true,
       "eu_egress_only": true,
       "engine_rule": {"deny": ["CN"]}}'

# Check what each key is allowed to do
curl https://api.specstream.ai/v1/account/keys -H "Authorization: Bearer <OAuth token>"

SpecStream is hosted in the EU and encrypted in transit and at rest, under a GDPR data processing agreement. No model is trained on your work.

A budget no run can pass

Token billing tells you the price after the run: Azure charges $10.00 per 1M input tokens and $40.00 per 1M output tokens for o3-deep-research, plus Grounding with Bing Search. SpecStream quotes the most a Deep Search run can cost, good for 30 minutes. The run starts only when that maximum is approved as confirm_max_cents, the charge never goes past it, and what the run does not use goes back to the balance.

Each key carries spend_cap_cents, the most it can spend in total. A platform that serves its own customers labels each call with X-End-User-Id and caps that customer for the month with X-End-User-Cap-Cents. A call past the cap gets 402 with end_user_cap.

curl "https://api.specstream.ai/v1/web-search?q=EU+AI+Act+obligations+for+deployers" \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" \
  -H "X-End-User-Id: customer-4812" \
  -H "X-End-User-Cap-Cents: 2000"

Every call on the record, for security and finance

The request log lists every Search API and Extract API call, charged or not, with its key, status, latency_ms and price_usd. Usage sums calls and spend by day, endpoint, key or end user, and exports as CSV. For a zero-retention key the log keeps only the time, endpoint, key, status, latency and price. The Enterprise plan, from $2,500 a month, adds single sign-on, SCIM, audit export and agreed terms.

# Every call, charged or not, with its key, status, latency and price
curl "https://api.specstream.ai/v1/account/requests?limit=100" -H "Authorization: Bearer <key>"

# Spend by day, endpoint, key or end user, as CSV for finance
curl "https://api.specstream.ai/v1/account/usage?group=key&format=csv" -H "Authorization: Bearer <key>"

Rows your systems can load

A research report still has to be read and typed into a table. Deep Search returns each match as a row with the columns you asked for, and every row meets the conditions you wrote. get_find_decisions returns each condition's answer with its reason, the value, the source and the time it was read. Duplicates, rows you already own and rows a page contradicts are not charged.

Source URL and read time on every value.

Volume prices and invoices

A search pack pays for Search API calls in advance and never expires: $5,000 buys 1,500,000 searches ($3.33 per 1,000) and $20,000 buys 8,000,000 ($2.50 per 1,000). On a signed order of 3 months or more, 5,000,000 searches cost $10,000 a month ($2.00 per 1,000). A top-up of $1,000 or more adds 20% more credit. To pay by invoice or sign for committed volume, talk to sales.

The search under it scores highest

The SpecStream Search API scores 97.9 F-score on all 1,000 official SimpleQA Verified questions (978 correct), and 64.0 item F1 and 43.4 row F1 on all 200 official WideSearch tasks. The Extract API scores 0.908 F1 on WCXB. Scores and method are on the benchmarks page.

Deep Research in Azure AI Foundry

OpenAI's o3-deep-research as an agent tool, announced in limited public preview for Azure AI Foundry Agent Service customers. It costs $10.00 per 1M input tokens and $40.00 per 1M output tokens, with Grounding with Bing Search and the clarifying model billed separately. SpecStream bills a delivered row at 5¢, 15¢ or $1, under a maximum approved first.

Gemini Deep Research

A preview agent that Google estimates at $1.00 to $3.00 a task, and $3.00 to $7.00 on Deep Research Max, at preview rates that can change. It researches for at most 60 minutes and does not support structured outputs. SpecStream returns structured rows, with the same row rates on every plan.

Tavily Research

A research endpoint that takes 4 to 250 credits a request, at $0.008 a credit pay as you go, so the charge lands between the minimum and the maximum. SpecStream names the most a run can cost before it starts.