Consensus API: Pricing, Features and Alternative

Choose SciSpace for a product that must deliver finished research files: extraction tables, cited reports and decks. Choose the Consensus API when you want paper retrieval as a separate component. Consensus also offers agent-written answers through Threads, so this is not simply search versus synthesis.

The SciSpace API is in early access; it isn't open yet. Fill out the Request SciSpace API access form to request access.

Last checked: October 28, 2026. This comparison uses official Consensus documentation and the capabilities of the SciSpace API in early access. No authenticated head-to-head requests were executed; the examples describe request contracts, not measured answers.

Consensus API vs SciSpace API at a glance

The decision is how much research orchestration your application should own. Compare the deliverable your users need, not just whether an endpoint returns text.

Consensus capabilities above follow its Search, full-text and Threads documentation. SciSpace capabilities describe its API in early access.

For your interface, distinguish a paper-results page from a completed research job. A research job needs an answer, inspectable references and any requested table or report, not just a list of promising papers.

What does the Consensus API cost per 1,000 calls?

As checked on October 28, 2026, Consensus lists additional usage at $0.05 per billable call on eligible plans. Therefore, 1,000 additional billable calls cost $50, excluding subscription fees and included allowances. This is not a quote for 1,000 research answers. Additional usage must be enabled. Pro and Teams each permit up to 500 additional calls per month; Deep permits the 1,000-additional-call example within its monthly maximum.

These details were checked against Consensus's plans and access documentation. For rate-limit errors, wait for the duration in retry-after before retrying. For allowance-related 429 responses, check the account's remaining allowance and the Thread's reserved usage instead of repeatedly retrying.

How should you test the same five research questions?

Use identical prompts and compare Consensus Threads against SciSpace Task API runs, rather than comparing a search list with a finished answer. Consensus documents Threads as its multi-step research interface.

The five questions below are an evaluation set, not fabricated findings. Both response columns remain unmeasured until authorized testing is possible.

Append this same instruction to each question:

Use evidence published through September 30, 2026. Give supporting and conflicting evidence, limitations and primary-paper citations. Include an evidence table with study design, sample or dataset, and outcome. Mark missing values as not reported.

Store that complete text in prompt for the Consensus example and send the same text as the SciSpace Task API request. Preserve prompts, settings and raw responses for review; a five-question check is not a general accuracy benchmark.

How do you send the questions to Consensus?

This shortened Python example retains the documented Threads request. Set prompt to one question plus the shared instruction, and repeat with the remaining questions. Authentication uses x-api-key.

Retain interaction_id too. Poll GET /v1/threads/{thread_id} using the returned polling guidance until idle or failed; the selected interaction carries the answer.

For retrieval alone, use GET /v1/search with query. That path returns ranked papers for your own reasoning pipeline.

What comes back from each research API?

For SciSpace, send the same question plus the shared instruction to the Task API as plain text. The Consensus column lists documented response locations; the SciSpace column describes what the Task API returns. Neither column contains captured answers to the five questions. Consensus's interaction structure follows its Threads guide.

Keep citations attached to their original answer. Preserve provider-specific payloads instead of flattening away evidence needed by reviewers.

Where does extraction fit?

For the battery question, ask SciSpace for chemistry, temperature, charging protocol and cycle-life definition as columns of a data-extraction table. The agent fills the columns from full text, and you can then filter, sort and combine the table.

SciSpace can also answer questions across up to 10 of your own PDFs at once, with page-level citations, and reads figures and tables. Consensus's API does not accept uploads of your own PDFs, as checked against its API docs on 28 Oct 2026.

Consensus Search excerpts can feed your own extractor, but they are query-relevant passages, not whole-paper exports. Excerpts require eligible open-access PDFs.

Which API is more accurate?

This article does not establish an accuracy winner. Workflow breadth is a reason to prefer an integration, not evidence that every generated claim is correct.

Review each answer for:

1. Relevance: Does the evidence match the population, material or task?
2. Citation support: Does the referenced passage support the claim?
3. Extraction fidelity: Are numbers, units and conditions preserved?
4. Disagreement: Are contradictory findings represented?
5. Uncertainty: Are missing values distinguished from negative results?

Use separate scores for retrieval and synthesis. For extraction, compare individual cells with the paper, not just the readability of the final table.

SciSpace writes cited reports with citation-checked sections, but that does not replace product-specific evaluation. For implementation context, see RAG API for research.

When should you choose Consensus instead?

Choose Consensus when you already own synthesis and want a focused retrieval component. Its Search API exposes structured filters including publication year, study type, human studies and sample size.

It also fits an existing pipeline that expects abstracts, paper metadata and relevant full-text excerpts. Test excerpt availability on your target literature rather than assuming every result includes passages.

Consensus Threads is a fair alternative for cited answers and citation crawling. Do not rebuild those capabilities unnecessarily; evaluate the additional deliverables your product requires.

How do you migrate from Consensus to SciSpace?

Treat switching as an adapter redesign, not a base-URL replacement. Consensus's request fields below follow its endpoint references; the SciSpace column describes the equivalent plain-language request.

Normalize answer text and citations, retain raw responses, and replay the evaluation set. The SciSpace agent does not stop mid-run to ask you questions, so put the scope and any screening criteria in the original task and use follow-ups in the same thread to adjust.

Where the SciSpace API fits next to Consensus

Consensus Threads stop at one Markdown answer: there is no streaming or webhooks (you poll about every 30 seconds), no file outputs, no upload of your own PDFs, and no trial, patent, grant or science-database search. Its systematic review is a recipe built on the API rather than a task. These points were checked against Consensus's live API docs on 28 Oct 2026.

The SciSpace API puts the full SciSpace Agent behind one call. It searches, reads, analyses and writes, and delivers the result as ready-to-use files.

What this adds on top of Consensus

• Finished files, many from one run. Markdown, Word, PowerPoint, a LaTeX report compiled to PDF with its BibTeX, posters and plots, instead of one text answer.
• Evidence beyond papers. One task can search 280M+ papers and also 11 national and regional clinical-trial registries plus the WHO registry, Google Patents, grant databases (Grants.gov, foundations, international funders) and 77 public science databases such as PubChem, ChEMBL, UniProt, ClinVar and PDB.
• Systematic review as a task. Research questions, screening of up to 3,000 papers, extraction, a PRISMA flow diagram and a report.
• Your own PDFs. Q&A over up to 10 PDFs at once with page-level citations.
• Live streaming. Each plan step and the answer stream as they are written, and follow-ups continue in the same thread.

When to use which

• Consensus when you want a bigger index (400M+), the deepest search filters (study type, sample size, journal quartile, clinical guidelines), a free API plan (30 calls a month), or ranked papers for a synthesis layer you already own.
• SciSpace when your product promises finished files, evidence beyond papers or visible progress while the agent works. For planning context, see the SciSpace API introduction.

What it doesn't do

• No paywalled papers. It works from your own PDFs plus 50M+ open-access full texts, so ask the agent to mark details it could not access.
• Smaller paper index and fewer search filters than Consensus. It searches SciSpace's 280M+ index plus Google Scholar, PubMed and arXiv, and does not query Semantic Scholar, Scopus, Web of Science or Lens.
• No mid-run questions. The agent does not stop to ask you anything while it runs; put scope and criteria in the task and use follow-ups to adjust.

The SciSpace API is in early access; it isn't open yet. At launch it's open to anyone: register on SciSpace and start, with no sales call. Every account gets $5 in free API credits, enough to try both APIs on real questions. The API is separate from SciSpace consumer plans; no consumer plan includes API usage.

Frequently asked questions

Sources

1. Consensus: Build with the Consensus API
2. Consensus API: Getting started
3. Consensus API: Plans and access
4. Consensus API: Search reference
5. Consensus API: Quick start for Search
6. Consensus API: Quick start for Threads
7. Consensus API: Full-text retrieval and excerpts
8. Consensus API: Start a Thread

注意转载授权 · 阅读原文