OpenAI has moved GPT‑Rosalind out of research preview and opened trusted-access applications globally for eligible life-sciences organisations. API billing begins on 5 October, while the separate Rosalind Workbench remains in research preview.

RELATED BUYER PROFILEReview GPT‑Rosalind pricing, access and test-pending evidencePricing · access · strengths · limitations →CONTINUE THE DECISIONBuild a source-led AI stack for research workEvidence · workflow · next action →CONTINUE THE DECISIONRun a same-task scientific model pilotEvidence · workflow · next action →CONTINUE THE DECISIONCalculate complete cost per accepted research resultEvidence · workflow · next action →
THE BRIEF IN 30 SECONDS

What you need to know

  • Eligible organisations can request GPT‑Rosalind through a qualification and safety review; availability is global but access is not automatic
  • API billing starts on 5 October 2026 at $5 per million input tokens, $0.50 per million cached input tokens and $25 per million output tokens
  • GPT‑Rosalind can be used through ChatGPT, Codex and the API by qualified customers; the separate Rosalind Workbench is still in research preview
THE NUMBERS

The release boundary

$5API input price per 1M tokens from 5 October
$0.50cached input price per 1M tokens
$25API output price per 1M tokens
50+tools and data sources in the free Codex research plugin
3access routes: ChatGPT, Codex and API
Pendingindependent same-task HubAI evaluation
01
THE CONTEXT

What changed on 11 September

OpenAI says GPT‑Rosalind has moved out of research preview and is now available globally to eligible organisations through its trusted-access programme. Qualified customers can use the specialised life-sciences model in ChatGPT, Codex and the API. The change widens the addressable market beyond the programme's earlier US-first enterprise framing, but it does not create instant self-service access: organisations still have to request access and pass qualification and safety review.

02
WHY IT MATTERS

GPT‑Rosalind pricing starts on 5 October

OpenAI lists the API identifier gpt-rosalind-research at $5 per million input tokens, $0.50 per million cached input tokens and $25 per million output tokens. Billing begins on 5 October 2026, and cache-write pricing does not apply. Those token rates are only the model layer. Web search, containers, specialist tools, data access, repeated runs and expert review can materially change the cost of an accepted research result.

03
WHAT HAPPENS NEXT

The model and Workbench are different products

GPT‑Rosalind is the specialist reasoning model. Rosalind Workbench is an orchestrated scientific workspace that connects tools and supports workflows such as sequencing analysis, evidence synthesis and experiment planning. OpenAI's current product page still labels the Workbench as research preview. Buyers should therefore record separate maturity, access and support assumptions for the model, the Workbench and any Codex-based workflow.

04
THE CONTEXT

What it is designed to do

OpenAI positions GPT‑Rosalind for biology, drug discovery and translational-medicine research, including literature and evidence synthesis, protein and variant analysis, molecular comparison, experimental planning and multi-step use of scientific databases. A freely available Life Sciences research plugin for Codex connects mainline models—and GPT‑Rosalind for eligible enterprise users—to more than 50 public tools and data sources. That plugin access does not itself grant GPT‑Rosalind access.

05
THE CONTEXT

Benchmark claims need the right label

OpenAI reports leading published performance on BixBench, improvements over GPT‑5.4 on six of 11 LABBench2 tasks and strong results in a Dyno Therapeutics evaluation. These are useful vendor-reported signals, not an independent HubAI test and not evidence of clinical effectiveness. A procurement team should reproduce the exact workflow it intends to buy, including source retrieval, tool calls, failed runs, scientific corrections and expert sign-off.

06
THE CONTEXT

HubAI buyer verdict

GPT‑Rosalind is a credible pilot candidate for governed research organisations that already have qualified scientific reviewers, controlled data access and a repeatable life-sciences workflow. It is not a shortcut around experimental validation, regulatory duties or research accountability. Before adoption, freeze one representative task, define an expert acceptance rule, compare it with the current model and process, and calculate total cost per accepted conclusion rather than token price alone.

HUBAI VIEW

Life-sciences teams should treat the release as a governed model evaluation, not a self-service launch: approval, scientific validation, tool traceability and complete research-task cost still determine fit.

Buyer decision signal: Expanded access · published API pricing
BUYER ACTION PLAN

What to verify next

1Confirm that the organisation and intended internal-research use can enter the trusted-access review

2Record the exact access route, model identifier and release date used in the pilot

3Separate GPT‑Rosalind model access from Rosalind Workbench and Codex plugin availability

4Define which scientific sources, databases, code and instruments the workflow can reach

5Require a qualified scientist to verify every material claim, calculation and proposed experiment

6Measure model tokens, tools, containers, retries and expert correction time per accepted result

7Review retention, data-processing region, access management and audit-log requirements

8Do not use vendor benchmark claims as proof of performance on the organisation's own data

Build a side-by-side comparison →
PRIMARY SOURCES

Read the evidence

Capabilities, availability and prices can change. HubAI keeps analysis separate from the underlying official material.

01OpenAI: Introducing GPT‑Rosalind for life sciences researchOpen source ↗02OpenAI API: GPT‑Rosalind pricingOpen source ↗03OpenAI: Rosalind product and Workbench accessOpen source ↗04OpenAI Developers: Rosalind WorkbenchOpen source ↗
Corrections & updates

Published and reviewed by the HubAI Intelligence Desk. Material product or policy changes are recorded with an updated timestamp.

Our editorial standard →