Tasks and inputs
GPT-Rosalind 6 is listed for Question answering, Text generation. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
GPT-Rosalind 6 (gpt-6-rosalind) is OpenAI's trusted-access reasoning model for biology, medicinal chemistry and genomics research by approved labs.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative, trusted access only. Up to 272K: $15 input / $1.50 cached / $75 output per 1M; cache writes $18.75 (30-min) / $37.50 (12-h). Over 272K: $30 / $3 / $112.50; cache writes $37.50 / $75. Confirm with your OpenAI account team.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Medicinal chemists do not need a model that writes nicer emails. They need one that can reason about a binding pocket, a synthesis route or a gene panel without wandering into confident nonsense. That is the audience for GPT-6 Rosalind, which OpenAI released on 30 October 2026 for complex life-sciences research, describing it as specialised biological reasoning with scientific tool use.
You apply for it. Access is limited to organisations approved for Rosalind Research through OpenAI's trusted-access programme, for internal research by authorised users.
Non-government applicants go through business verification and compliance screening. Government bodies working on biosecurity, biosafety, public-health preparedness or biodefence have a separate pathway. Named early partners include Amgen, Moderna, Novo Nordisk and Thermo Fisher.
There is a condition attached that procurement teams should read twice: enterprise users must apply qualified human review before outputs inform experiments, publications or decisions. That is a contractual workflow requirement, not a suggestion.
It runs on the same weights as GPT-6 Astra. There is no separate training run; the difference lies in deployment configuration, safeguards and access controls, including a refusal boundary adjusted for legitimate biology work. OpenAI rates it High, below Critical, for biological and chemical risk, while cybersecurity stays at Critical, inherited from Astra.
That base-model detail, and the benchmark figures below, rest on a single second-hand account of OpenAI's safety material. OpenAI has not published a model page that confirms them. Treat them as strong context, not a specification.
| Band | Input | Cached input | Output | Cache writes (30-min / 12-h) |
|---|---|---|---|---|
| Up to 272K | $15 | $1.50 | $75 | $18.75 / $37.50 |
| Over 272K | $30 | $3 | $112.50 | $37.50 / $75 |
All per million tokens, and indicative. That is 1.5 times Astra's standard rate. Trusted-access contracts may differ, so confirm with your account team.
The naming is untidy. OpenAI's model list shows "GPT-Rosalind" with no ID. Its price list and release notes use gpt-6-rosalind. Elsewhere it appears as "GPT-Rosalind-6". All three refer to this model; call it by the API ID.
Two other SKUs sit beside it and are easy to confuse:
gpt-rosalind-research, generally available to approved organisations since 8 September 2026, billed from 5 October 2026 at $5 input, $0.50 cached and $25 output per million tokens.gpt-rosalind-discovery, also listed at $5 input and $25 output.Neither carries long-context pricing. If your work fits in a short context and does not need the frontier reasoning tier, those are a third of the price.
OpenAI's internal comparisons against GPT-Rosalind-5.5 show MedChemBench v7 rising from 28.9% to 49.6%, GeneBench Pro v13 from 16.7% to 37.7%, and LifeSciBench Gold from 44.0% to 61.3%. It reportedly scored 0.751 pass@1 on BixBench. These are vendor figures on vendor-built suites; nobody outside OpenAI has verified them.
Input and output modalities, context window and maximum output are not published on a first-party page. Tool support is described only as "scientific tool use". Before committing, ask for the model card and run your own holdout set: twenty questions from past projects where you already know the answer, scored by the chemist or biologist who solved them. Measure how often the model is right, how often it says it does not know, and how long review takes. The review cost is part of the price.
General reasoning models without the access gate are listed on the OpenAI provider page and under question answering.
Compare the model on the work you need to complete.
GPT-Rosalind 6 is listed for Question answering, Text generation. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
The listed route has starting rates of $15.00 input and $75.00 output per million tokens. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Microsoft
Aion 1.0 Instruct 2.3B is Microsoft's MIT-licensed on-device small language model and Phi Silica successor for Windows, with a 32K context.
Gemini 4 Neon is Google's mid-tier Gemini 4 model for complex coding and analysis, sitting between Hydrogen and Carbon at $3/$15 per 1M tokens.
OpenAI
GPT-6.2 Astra is OpenAI's November 2026 flagship for hard professional, scientific and agentic work, at $8/$40 per million tokens with a 1.05M context.
OpenAI
GPT-6.2 Sol is OpenAI's everyday model for coding, agents and writing: $2/$10 per million tokens, 1.05M context, released 17 November 2026.
Alibaba
Qwen4 Max (qwen4-max) is Alibaba's first Qwen4-generation flagship: 1M-token context, text, image and video input, thinking on by default, $1.50/$4.50 per 1M.
Sarvam
Sarvam-105B-1711 is a refreshed open post-train of Sarvam's 105B MoE, tuned for tool calling, instruction following, agents and Indian languages.