Skip to the page
Get started free

Azure o3-deep-research retires Nov 19, 2026

Where to move.

Microsoft retires o3-deep-research version 2025-06-26 in Azure OpenAI on November 19, 2026. Its replacement is gpt-5.6-sol, a general model you pair with web search. The classic Deep Research tool in Foundry Agent Service goes the same day.

SpecStream is the live web data engine for teams and agents. POST /v1/research takes the same research prompt and returns a written report with the page behind each cited sentence, at $0.10, $0.30 or $1.00 a report.

Get an API key

New accounts get $10 of credit, no card. It covers 100 quick reports or 33 standard reports.

SpecStream, o3-deep-research on Azure and its replacement

SpecStreamAzure o3-deep-researchAzure gpt-5.6-sol with web search
From Nov 19, 2026POST /v1/research answers, on the same key as the Search API and the Extract APIRetired. Microsoft's policy: a retired model is removed from service and all inference requests return 410 GoneGenerally available, version 2026-07-09, with a retirement date of January 11, 2028
Who can start todayAnyone. New accounts get $10 of credit, no cardExisting customers. The lifecycle is Deprecated, so a subscription that never deployed this version cannot create a deploymentAny Azure customer. A generally available model takes new deployments
The callPOST https://api.specstream.ai/v1/research with input, then GET /v1/research/{id}The Responses API with model set to your deployment, background: true and at least one data source: web_search_preview or a remote MCP serverMicrosoft recommends the latest Agents GA API with the latest GPT models and web search
What comes backreport.content, the text with [1] markers. report.citations gives each cited sentence its page, read time and passage. report.sources lists every page readA citation-rich report: a message with inline url_citation annotations, after the list of web_search_call itemsA grounded response with citations
Where the web search runsSpecStream reads the live web for each report. Source URL and read time on every citationGrounding with Bing Search. Regions: West US and Norway EastGrounding with Bing Search. Microsoft says data transfers occur outside compliance and geographic boundaries
Price$0.10 quick, $0.30 standard, $1.00 deep. Taken once, when the report is delivered. A cancelled report is not charged$10.00 input and $40.00 output per 1M tokens, Global list price on a tracker that reads the Azure Retail Prices API. Grounding with Bing Search is billed on top$8.00 per 1M tokens blended at 3:1 input to output on the same tracker, against $17.50 for o3-deep-research. Grounding with Bing Search is billed on top
Price before the runShown as price_usd in the reply to the POST, one price for each depthMetered as the run goes. Microsoft's advice is max_tool_calls to control tool usage and manage cost and latencyMetered by tokens, and use of Grounding with Bing Search incurs costs

What retires on November 19, and why teams move to SpecStream

What ends on November 19

Microsoft's model retirement schedule lists o3-deep-research version 2025-06-26 as Deprecated with the retirement date 2026-11-19 and the replacement gpt-5.6-sol. The same date applies to o1, o1-pro, o3, o3-pro, o3-mini and o4-mini. Microsoft's lifecycle policy says a retired model is removed from service and all inference requests return 410 Gone.

The schedule also says the Replacement column is, where supported, the target for automatic upgrades at retirement. For Global Standard, Data Zone Standard and Standard deployments Microsoft manages those upgrades, and o3-deep-research deploys as Global Standard. So plan for one of two results: a 410, or a deployment that now answers as gpt-5.6-sol, a general model. Provisioned deployments are not upgraded for you.

The classic Deep Research tool in Foundry Agent Service ends with the model. Microsoft's page says the tool will be deprecated when the o3-deep-research model retires on November 19, 2026.

The replacement is a general model with a search tool

o3-deep-research was built for one job: browse, analyze and synthesize hundreds of sources into a citation-rich report. gpt-5.6-sol is a general model. On Azure you give it the web search tool, write the research instructions yourself and test the reports again. Microsoft's own advice is to evaluate available models with your application and data, comparing quality, latency and cost.

A SpecStream research report is the same job as one call. You send the question as input and choose quick, standard or deep. A quick report takes about a minute, a standard report about four minutes and a deep report about five.

Each cited sentence comes with its passage

An Azure deep research answer carries url_citation annotations with a title, a URL and character offsets. A SpecStream report goes one step further. report.citations gives each cited sentence its page, the time it was read and the passage that supports it, and report.sources lists every page read in the run. Your reader checks a sentence against the passage without opening the whole page. The research report docs list every field.

You know the price before the report starts

On Azure a research run is metered by tokens and tool calls as it goes, and Grounding with Bing Search is billed on top. A SpecStream report costs $0.10 for quick, $0.30 for standard and $1.00 for deep. The reply to the POST shows the price as price_usd, the balance pays it once when the report is delivered, and a cancelled report costs nothing. See pricing.

No deployment, no region and no Bing resource

The Azure setup needs a model deployment in West US or Norway East, and the classic tool also needs a gpt-4o deployment and a Grounding with Bing Search resource in the same subscription and region. SpecStream needs one API key. For JSON in place of text, send "output": {"type": "json", "schema": {...}}. include_domains and exclude_domains choose the sites, and a webhook is called when the report ends.

Search and page reading on the same key

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. The Search API costs $5 per 1,000 searches of up to 10 results, or $1 with speed=fast, and the Extract API costs $0.80 per 1,000 pages.

What else Azure deep research users price

OpenAI's own API

SpecStream writes the report for $0.10, $0.30 or $1.00. The same model is gone from OpenAI's API: OpenAI shut down o3-deep-research on July 23, 2026, and its deprecations page names gpt-5.6-sol there too. The OpenAI deep research page has the details.

Gemini Deep Research

Two preview agents on Google's Interactions API. Google estimates $1.00 to $3.00 a task, and $3.00 to $7.00 on Deep Research Max, at preview rates that can change, and its guide says the agent does not support structured outputs. A SpecStream deep report costs $1.00 and returns JSON when you send a schema.

Parallel Task API

Priced per run by processor, from $5 per 1,000 runs on lite to $2,400 per 1,000 on ultra8x. The ultra processor costs $300 per 1,000, or 30¢ a run, and takes 5 to 25 minutes. A SpecStream standard report costs the same 30¢ and takes about four minutes.

How SpecStream scores

The same numbers as the benchmarks page, from the official tasks and scorers.

SpecStreamParallelExaTavilyPerplexity
Web researchWideSearch, GPT-5.6 Luna60.945.553.047.947.0
Factual accuracySimpleQA Verified, GPT-5.6 Luna9694919494
Factual accuracySimpleQA Verified, GPT-5.6 Sol9997919295

How to move an Azure o3-deep-research call before Nov 19

1. Find every call. Search for your o3-deep-research deployment name in responses.create calls, and for the classic Deep Research tool in Agents SDK code on the 2025-05-15-preview API.

2. Create a SpecStream key under Agents & API in the workspace. New accounts get $10 of credit, no card.

3. Start the report. Your research prompt becomes input, up to 15,000 characters, and depth takes the place of max_tool_calls. The reply is 202 with the report's id and its price_usd.

curl -X POST https://api.specstream.ai/v1/research \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input": "Research the economic impact of semaglutide on global healthcare systems.", "depth": "standard"}'

4. Read it until status is completed, as you polled the background response. report.content takes the place of the final message, and report.citations the place of its url_citation annotations.

curl https://api.specstream.ai/v1/research/REPORT_ID \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY"

5. If you used a webhook on Azure, pass "webhook": {"url": "https://..."}, or follow GET /v1/research/{id}/events. Send an Idempotency-Key header on the POST so a retry returns the same report and is charged once.

To stay on Azure, deploy gpt-5.6-sol and give it SpecStream as a remote MCP server in the Responses API. The model then searches and reads the live web through SpecStream, at $5 per 1,000 searches of up to 10 results and $0.80 per 1,000 pages, with no Bing resource.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/",
    api_key=os.getenv("AZURE_OPENAI_API_KEY"),
)

response = client.responses.create(
    model="gpt-5.6-sol",  # your deployment name
    background=True,
    tools=[{
        "type": "mcp",
        "server_label": "specstream",
        "server_url": "https://api.specstream.ai/mcp",
        "headers": {"Authorization": "Bearer " + os.environ["SPECSTREAM_API_KEY"]},
        "require_approval": "never",
    }],
    input="Research the economic impact of semaglutide on global healthcare systems.",
)
print(response.id, response.status)

The same MCP server runs the whole report for an agent: research_price shows the price of each depth, start_research starts the report once the person who pays approves that price, and get_research reads it. The MCP guide has the setup for each client.

Get an API key

Frequently asked questions

Something we didn't answer? Talk to a human

When does o3-deep-research retire on Azure?
November 19, 2026. Microsoft's model retirement schedule lists o3-deep-research version 2025-06-26 in Azure OpenAI as Deprecated, with the retirement date 2026-11-19.
What replaces o3-deep-research on Azure?
Microsoft names gpt-5.6-sol, version 2026-07-09, which it lists as generally available until January 11, 2028. It is a general model, so the research behavior comes from the web search tool and your own instructions.
What happens to my deployment on November 19?
One of two things. Microsoft's policy says all inference requests to a retired model return 410 Gone, and its schedule says the replacement is, where supported, the target for automatic upgrades at retirement. Provisioned deployments are not upgraded automatically.
Is the Deep Research tool in Foundry Agent Service affected?
Yes, on the same day, November 19, 2026. Microsoft's page says the tool will be deprecated when the o3-deep-research model retires, and recommends the latest Agents GA API with the latest GPT models and web search.
Can I use o3-deep-research on OpenAI's API in place of Azure?
No. OpenAI shut the model down on July 23, 2026 and names gpt-5.6-sol as the replacement.
What does a SpecStream research report cost?
$0.10 for quick, $0.30 for standard and $1.00 for deep. The price is taken once, when the report is delivered, and a cancelled report is not charged. New accounts get $10 of credit, no card.
Does SpecStream have an API and an MCP server?
Yes, and new accounts get $10 of credit, no card. The report runs at POST https://api.specstream.ai/v1/research, and over MCP at https://api.specstream.ai/mcp with research_price, start_research and get_research.

Sources

SpecStream wrote this page. Each Microsoft fact comes from Microsoft's own pages, and the Azure token prices from a public tracker that reads the Azure Retail Prices API.

  1. SpecStream pricingspecstream.ai/pricing, read Nov 13, 2026
  2. SpecStream for agentsspecstream.ai/agents.md, read Nov 13, 2026
  3. SpecStream benchmarksspecstream.ai/benchmarks, read Nov 13, 2026
  4. Microsoft Foundry model retirement schedulelearn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule, read Nov 13, 2026
  5. Microsoft Foundry Models lifecycle and support policylearn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirements, read Nov 13, 2026
  6. Azure OpenAI deep research with the Responses APIlearn.microsoft.com/en-us/azure/foundry/openai/how-to/deep-research, read Nov 13, 2026
  7. Foundry Agent Service Deep Research tool (classic)learn.microsoft.com/en-us/azure/foundry-classic/agents/how-to/tools-classic/deep-research, read Nov 13, 2026
  8. Foundry web search toollearn.microsoft.com/en-us/azure/foundry/agents/how-to/tools/web-search, read Nov 13, 2026
  9. Azure OpenAI Responses APIlearn.microsoft.com/en-us/azure/foundry/openai/how-to/responses, read Nov 13, 2026
  10. Foundry model availability tracker: o3-deep-researchjinlee794.github.io/foundry-model-availability-notifications/models/o3-deep-research/, read Nov 13, 2026
  11. OpenAI o3-deep-researchdevelopers.openai.com/api/docs/models/o3-deep-research, read Nov 13, 2026
  12. OpenAI deprecationsdevelopers.openai.com/api/docs/deprecations, read Nov 13, 2026
  13. Gemini Deep Research agentaistudio.google.com/docs/deep-research, read Nov 13, 2026
  14. Parallel pricingdocs.parallel.ai/getting-started/pricing.md, read Nov 13, 2026