NewAnswers MCP: one question, one checked answer for your agent.See how it works

Alternatives

The OpenAI deep research API after the July 2026 shutdown

By , founder of HyperresearchUpdated

OpenAI no longer sells a deep research model through its API. Its deprecations page says access to o3-deep-research and o4-mini-deep-research “was shut down on July 23, 2026” and lists gpt-5.6-sol as the replacement for both. Deep research on OpenAI is now a way of calling a general reasoning model: the Responses API, the web_search tool, high reasoning effort and background mode. You pay per token plus $10.00 per 1,000 web search calls, so the price of a report depends on how long the model searches and reads. If you would rather buy a research task as one call at a known price, the alternatives are below.

We make Hyperresearch, one of those alternatives, so read this as a vendor’s guide. Every OpenAI fact on this page comes from OpenAI’s own docs, read on November 3, 2026, and is linked where it is used. For a side-by-side of every research API that returns citations, see deep research APIs with citations, compared.

What OpenAI shut down and what its docs name now

On April 22, 2026, OpenAI announced the deprecation of a set of older models, including both deep research models and their -2025-06-26 snapshots. They stopped working on July 23, 2026. The deprecations table lists gpt-5.6-sol as the substitute for each.

OpenAI’s other pages do not all name that model:

  • The deep research guide opens with the shutdown notice and keeps its old examples “for reference”. It adds that “changing the model ID alone is not a complete migration” and points to the Responses API, web search, file search and remote MCP guides.
  • The web search guide describes deep research as an extended investigation by a reasoning model that “can run for several minutes and is best used with background mode”. Its text says to use gpt-5.5 with reasoning set to high or xhigh.
  • The same guide’s example for longer research runs uses gpt-6-astra at xhigh reasoning, with the web_search tool’s return_token_budget set to unlimited.

So there is no single deep research model id any more. gpt-5.6-sol is the listed replacement, and its model page shows xhigh reasoning and the web search, file search, code interpreter and MCP tools in the Responses API.

OpenAI deep research pricing now

The pricing page has no row for the retired deep research models. A research call is billed as tokens plus tool calls at these rates, per million tokens, standard processing:

Model Input Cached input Output Note
gpt-5.6-sol $4.00 $0.40 $20.00 The listed replacement. Promotional pricing “at least through November 21, 2026”
gpt-5.5 $5.00 $0.50 $30.00 The model the web search guide’s text names
gpt-6-astra $10.00 $1.00 $50.00 The model in the guide’s long-research example

Prompts over 272K input tokens bill at long-context rates for the whole request; for gpt-5.6-sol that is $8.00 in and $30.00 out. Batch and Flex processing list gpt-5.6-sol at $2.00 in and $10.00 out.

Tool calls add:

  • Web search: $10.00 per 1,000 calls, plus the search content fed to the model, billed as input tokens at the model’s rates.
  • File search: $2.50 per 1,000 calls, plus $0.10 per GB a day of storage after the first GB.

An example, with assumed token counts rather than measured ones: a research call on gpt-5.6-sol that makes 50 searches, reads 1 million input tokens and writes 50,000 output tokens costs, by our arithmetic:

50 searches x $10.00 / 1,000 = $0.50
1,000,000 in x $4.00 / 1M = $4.00
50,000 out x $20.00 / 1M = $1.00
-----
$5.50

If that input arrives in requests over 272K tokens, the long-context rates apply and the same work comes to about $10.00. Reasoning tokens bill as output, and return_token_budget: "unlimited" lets the search tool return more content, which the guide warns “can increase latency and cost”. You only know the real figure after the response finishes.

Migrating an o3-deep-research call

The old call was one Responses API request with a research model and at least one data source:

POST https://api.openai.com/v1/responses
{
"model": "o3-deep-research",
"input": "<question>",
"background": true,
"tools": [{ "type": "web_search_preview" }]
}

The nearest call today, assembled from the web search guide’s long-research example with the model set to the listed replacement:

POST https://api.openai.com/v1/responses
{
"model": "gpt-5.6-sol",
"input": "<question>",
"reasoning": { "effort": "xhigh" },
"background": true,
"tools": [{ "type": "web_search", "return_token_budget": "unlimited" }]
}

What changes when you move:

  • The tool. return_token_budget and the hosted search controls (domain filters, sources) apply to web_search, not web_search_preview.
  • Waiting. Background mode still works: poll GET /v1/responses/{id} while the status is queued or in_progress, or cancel with POST /v1/responses/{id}/cancel.
  • Citations. Inline citations come back as url_citation annotations with the URL, title and position in the text. The sources field lists every URL the model consulted, which is often more than it cited.
  • Your own documents. file_search over vector stores and remote MCP servers are still available as tools.
  • The research behaviour. The deep research models had their own way of breaking a question down and deciding when to stop. With a general model, your prompt and reasoning effort set that, and checking that each citation supports its sentence is still your job.

Research task APIs from other vendors

Some vendors sell a research task as one call, at a price you can see before you run it or close to it:

  • Parallel Task API. Priced per run, not per token. The ultra processor, “Advanced multi-source deep research”, is $300 per 1,000 runs, or $0.30 a task, and ultra8x is $2,400 per 1,000. Failed runs are not billed (pricing).
  • Perplexity Agent API. Perplexity has retired Sonar, including sonar-deep-research. The high effort level bills tokens plus $0.0025 per web search and $0.0005 per URL fetch, and cost $0.12 to $0.14 a request on Perplexity’s own benchmarks.
  • Valyu DeepResearch. A fixed price per task in credits, with searches and page reads included: 100 credits for standard, or about $0.63 at the pay-as-you-go rate of 1,600 credits per $10 (our arithmetic). The Valyu comparison has every mode.

These keep the one-call shape of the old API, at prices from about $0.12 a request to $2.40 a task.

Moving to Hyperresearch

A Hyperresearch run takes one question and returns a long report, the sources behind it and the claims it relies on; a Deep run adds a verification receipt. The receipt lists the sampled citations, each checked against the source it cites, with a verdict of supported, partially supported, unsupported or wrong source. Retracted papers and quotes that do not appear in their source stop a report from shipping.

Two sizes, flat per run:

  • Light, $9: about 20 to 50 sources and 2,500 to 4,000 words, in about 30 to 40 minutes.
  • Deep, $49: about 150 to 200 sources, split into 3 to 5 chapters, in 3 to 5 hours.

Source counts and times are the targets the pipeline gates on, not guarantees. The price is fixed when the run is created, not metered by tokens.

On ten DRACO tasks, Hyperresearch scored 77.7

The retired API models can no longer be tested, so the nearest OpenAI product we could grade is Deep research in the ChatGPT app. We ran it, Hyperresearch Deep, Perplexity’s sonar-deep-research API and other research agents on the same 10 tasks from DRACO, a deep research benchmark from Perplexity and Harvard, in October 2026, and graded each with the same judge, GPT-5.2 at the paper’s settings, mostly as the mean of two grades. Valyu DeepResearch Heavy, a research task API, was graded once from the reports Valyu published for DRACO.

System DRACO tasks Mean score
Hyperresearch Deep 10 of 10 77.7
Perplexity Deep Research (API) 8 of 10 57.8
Valyu DeepResearch Heavy 8 of 10 57.2
ChatGPT Deep research (app) 10 of 10 55.2

Against ChatGPT the widest gaps are on the Finance task, 70.8 against 27.8, and the UX design task, 74.2 against 33.8. ChatGPT came closest on the Shopping task, 72.4 against 80.2. Valyu and Perplexity scored higher than us on the Needle in a haystack task. ChatGPT’s reports were exported from the app without source links, so its citation scores reflect that. Perplexity’s and Valyu’s means cover the 8 tasks with a graded report. These are 10 of DRACO’s 100 tasks, two grades of the same report can differ by up to about 9 points, and this is our measurement, not an official leaderboard. How we tested has every task score and the other systems we graded, including Claude Research (Opus) at 73.0, or see a full Deep report.

The shapes side by side

OpenAI Responses API with web search Research task API Hyperresearch
You pay for Tokens plus search calls Per task, or tokens plus searches Per run, $9 or $49
Price known After the response finishes Before the run, or after for token-billed APIs Before the run starts
Research loop Your prompt and reasoning effort The vendor’s The pipeline’s
Citations url_citation annotations with positions, plus sources Varies by vendor Every kept source in your vault, with a verification receipt on Deep
Citation checking Yours to build Varies by vendor Built in
Time per call Several minutes or more Seconds to hours by processor About 30 to 40 min Light, 3 to 5 h Deep

A $0.30 task and a $49 run are different purchases. Decide which one you need before comparing prices.

The call, field by field

Terminal window
curl -X POST https://api.hyperresearch.ai/v1/runs \
-H "Authorization: Bearer <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: migrate-0001" \
-d '{
"query": "<question>",
"tier": "premier",
"levers": { "register": "analyze", "purpose": "decide" },
"webhook_url": "https://example.com/hooks/hyperresearch"
}'
OpenAI Hyperresearch Note
input query Up to 20,000 characters
model and reasoning.effort tier light or premier. premier is Deep on the wire. Omitted, it is light
background: true nothing Every run is asynchronous. The POST answers 202
tools: [web_search] nothing Searching and fetching are part of the run
file_search with vector_store_ids your workspace vault, and project_id Runs read the vault before the web. A project_id scopes that reuse to one project
polling the response id GET /v1/runs/{id}, or webhook_url Use either
tone and structure in the prompt levers register, purpose, reader_expertise, report_language, response_format, required_section_headings, domain_notes
none Idempotency-Key Same key and body returns the original run. Same key with a different body is a 409

The 202 gives you the run id and the price before any work starts:

{
"id": "run_01J9ZQ7V6H0000000000000000",
"status": "queued",
"tier": "premier",
"price_cents": 4900,
"events_url": "...",
"created_at": "2026-09-18T14:02:11Z"
}

Waiting for the run

Poll GET /v1/runs/{id} for the status, the current step, steps_done out of steps_total, sources kept so far, and on a failed run a failure_reason. A run can also stop at blocked and wait for a decision, for example a source it could not read without your help. blocked_code tells you which gate it stopped on.

To skip polling, set webhook_url. Events include run.started, run.step, run.blocked, run.done, run.failed and verify.done.

When the run is done, GET /v1/runs/{id}/result returns report_markdown, sources, skipped (every page tried and not kept, with a reason), claims, verification, escalations and artifacts. Artifact links are signed URLs valid for one hour.

If an agent is calling rather than a backend, the same start, check and fetch pattern is available over MCP at https://mcp.hyperresearch.ai/mcp. Long-running MCP tools explains why it takes three calls.

When OpenAI or a task API is the better choice

Pick OpenAI if your app already runs on the Responses API, you want token-level control of the research, and your costs per call land in the single dollars or below. The estimate above came to $5.50 for a fairly heavy call. You also keep one vendor and position-level citations.

Pick a task API if you need an answer while a user waits, a price per task you can quote in advance, or thousands of lookups a day. A Hyperresearch run takes about half an hour or more and costs at least $9. No amount of checking makes that the right purchase for a quick lookup.

Pick Hyperresearch when the report is the thing you deliver and someone will be held to what it says. There, the difference between a citation that exists and one that supports its sentence is worth paying for. Pricing has the full numbers.