FrontierScience is live · free · last verified 26 Oct 2026 · viability monitor
FrontierScience
OpenAI's December 2025 wet-lab evaluation where GPT-5 improved a real molecular cloning protocol 79-fold
Read FrontierScience as a capability and biosecurity signal, not as lab software. The headline isn't 79x by itself — it's that RecA plus gp32 was not in the starting protocol and the gain survived sequencing with n=3 error bars, meaning GPT-5 proposed a genuinely new wet-lab mechanism. Keep the caveats loud: one cloning protocol, one benign GFP/pUC19 system, fixed prompting, and humans still running every reaction. If you need protocol optimization on your own assay today, this publication gives you a method to copy, not a service to buy; if you're assessing frontier-model risk in biology, it's required reading.
Verdict generatedVerified 4d ago · liveness 60/100 · cite: rightaichoice.com/tools/frontierscience
- AI researchers evaluating frontier model reasoning in experimental science
- Scientific institutions assessing whether AI can improve real wet lab protocols
- Biosecurity analysts working under a Preparedness-style risk framework
- Molecular biologists interested in RecA/gp32 homology-based cloning improvements
- Teams without a wet lab, reagents, and scientists to execute protocols
- Researchers who need protocol optimization across many different assays
- Anyone expecting a purchasable product rather than a published evaluation
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip FrontierScience if you need to optimize a protocol on your own assay this quarter — the publication gives you the method and the evidence, but no hosted environment, and running the loop yourself means lining up a wet lab and the Red Queen Bio collaboration.
The real cost is bench time, not software: every optimization round needs scientists to set up and run reactions and upload colony counts for 8-10 candidates at a time
There is no price to weigh: the work is a free, openly readable OpenAI research publication dated December 16, 2025, not a subscription or per-seat tool. Compare it with what a paid platform would cost you instead — ChatGPT plans sit at the individual level on openai.com/pricing, and lab automation suites are enterprise contracts. FrontierScience's cost shows up as reagents, sequencing, and scientist hours.
In short
FrontierScience — OpenAI's December 2025 wet-lab evaluation where GPT-5 improved a real molecular cloning protocol 79-fold. Best for AI researchers evaluating frontier model reasoning in experimental science, Scientific institutions assessing whether AI can improve real wet lab protocols, Biosecurity analysts working under a Preparedness-style risk framework. Free to use.
What is FrontierScience?
FrontierScience — OpenAI's December 2025 wet-lab evaluation where GPT-5 improved a real molecular cloning protocol 79-fold. Best for AI researchers evaluating frontier model reasoning in experimental science, Scientific institutions assessing whether AI can improve real wet lab protocols, Biosecurity analysts working under a Preparedness-style risk framework. Free to use.
FrontierScience is OpenAI's name for a December 16, 2025 research publication and evaluation framework, built with biosecurity startup Red Queen Bio, that tests whether a frontier model can propose, analyze, and iterate on genuine wet-lab experiments rather than recite known biology. There is no login and no sign-up: the deliverable is a published report you read, not software you run. The task was a two-piece HiFi/Gibson molecular cloning reaction using GFP and the pUC19 plasmid, picked for a clean optimization signal and 1-2 day experimental cycles. GPT-5 reasoned about the protocol, proposed batches of 8-10 candidate reactions per round, and revised them from colony-count data that human scientists produced by actually running the reactions and uploading results. Prompting was fixed, with no human guidance on the model's reasoning. Across rounds GPT-5 introduced a mechanism absent from the starting protocol: E. coli RecA recombinase paired with phage T4 gp32 single-stranded DNA-binding protein, with gp32 detangling loose DNA ends before RecA guides each strand to its match. Screening split the wins into RecA-Assisted Pair-and-Finish HiFi Assembly (RAPF, 2.6-fold) and Transformation 7 (T7, 36-fold), which combined to 79x more sequence-verified clones from a fixed amount of input DNA, with error bars from n=3 independent validation experiments. The fixed-prompt scaffolding also locked the loop into exploration and capped its ability to exploit its own discoveries; OpenAI says a better exploration/exploitation balance would likely yield larger gains. Scope matters as much as the number: benign system, deliberately narrow task, improvements specific to this cloning setup, humans still pipetting every reaction, and a closed robotic lab loop described as a future direction rather than demonstrated. Read it as a capability and biosecurity signal under OpenAI's Preparedness Framework, not as lab software you can adopt.
Key features
- Autonomous AI-lab loop with fixed prompting and no human guidance on model reasoning
- Model proposes batches of 8-10 candidate reactions per round and revises from new data
- Online learning from prior-round colony count feedback uploaded by human scientists
- Novel mechanism discovery pairing E. coli RecA recombinase with phage T4 gp32
- Optimization of two-piece HiFi/Gibson molecular cloning protocols
- 79x gain in sequence-verified clones versus the baseline HiFi reaction
- Gains split into RAPF assembly (2.6x) and T7 transformation (36x)
- Quantitative results with error bars from n=3 independent validation experiments
- Benign experimental system using GFP and the pUC19 plasmid
- Built with biosecurity startup Red Queen Bio
- Feeds biosecurity risk assessment under OpenAI's Preparedness Framework
- Tests hypothesis generation, revision, and mechanistic reasoning
- Measures model reasoning across physics, chemistry, and biology
- Requires human scientists to execute protocols and upload experimental data
- Robotic closed-loop system integration described as a future direction, not demonstrated
Behind the verdict
The interesting part of this work is the loop structure, not the multiplier. OpenAI and Red Queen Bio built an evolutionary scaffold: GPT-5 proposes 8-10 candidate reactions, scientists execute them, colony counts come back, and the model revises. Fixed prompts and no human steering of the model's reasoning mean any novel insight belongs to the model rather than to a clever operator — which is exactly what makes the RecA/gp32 discovery worth noting. gp32 detangles loose DNA ends, RecA then guides each strand to its correct match, and the effect shows up as a 2.6-fold RAPF assembly gain on top of a 36-fold T7 transformation gain, combining additively to 79x more sequence-verified clones per unit of input DNA. Strengths: sequencing-verified rather than narrative results; a scoped, benign experimental system; error bars from independent validation; and an honest account of the framework's own limits. OpenAI states plainly that fixed prompting locked the system into exploration and limited exploitation of what it found, and that both the enzymatic and transformation improvements have substantial room to refine — a candor you rarely get in launch write-ups. Where it falls short for a buyer: there is nothing to sign up for, no API, no hosted environment, and no path to point this loop at your own assay other than working with the authors. The result is specific to one GFP/pUC19 HiFi cloning setup, so protein engineering teams or strain-engineering groups should not assume transfer. Robotic closed-loop integration appears only as a future direction. Where it fits: AI evaluation teams comparing autonomous discovery loops against static Q&A benchmarks; scientific institutions judging whether frontier models can improve a real protocol; biosecurity analysts applying a Preparedness-style framework. Compare it to how you'd read a benchmark paper rather than how you'd evaluate a lab-automation vendor — the method is the takeaway, and replication on a second assay is what would move this from signal to evidence.
Models under the hood
as of 2026-10-24
Limitations
- This is an OpenAI research evaluation published December 16, 2025 with Red Queen Bio, not a product you can adopt, and the public evidence covers a single demonstration: GPT-5 optimizing one molecular cloning protocol.
- The system used was simple and benign (GFP and the pUC19 plasmid) with fixed prompting and no human guidance on the model's reasoning, and the 79x gain is specific to that setup.
- Human scientists were still required to execute every protocol and upload colony counts, and robotic closed-loop integration is described only as a future direction.
- Fixed prompting also locked the loop into exploration, limiting exploitation of its own discoveries, so larger gains are plausible but not yet shown.
as of 2026-10-26
Researching FrontierScience? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is FrontierScience still active?
FrontierScience is live · free · last verified 26 Oct 2026 · viability monitor.
Viability score
How well maintained and how widely used is FrontierScience? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →What do users say about FrontierScience?
What people actually say about FrontierScience — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
ResearchedDiscussed by users across 2 sources (Hacker News, YouTube) · researched Aug 11, 2026.
Average across the 2 sources that answered — each source counts once, not each post.
- +Real wet lab experiments, not just static Q&A.
- +GPT-5 demonstrated 79x efficiency gain with novel enzymes.
- +Controlled biosecurity settings for safe evaluation.
- +Collaboration with Red Queen Bio adds domain credibility.
- +Measures hypothesis generation and iterative reasoning.
- −Requires advanced lab infrastructure, not accessible to most.
- −Hard to verify claims without independent replication.
- −Visualization inconsistencies undermine data credibility.
- −Community discussion is often shallow or off-topic.
- −Limited user guides and examples for setup.
- • Requires specialized lab equipment and consumables not covered
- • Robotic system integration costs and maintenance not included
- • Potential need for professional biosecurity clearance in some jurisdictions
How we verified this
Every listed fact about FrontierScience— pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 26 Oct 2026. Pages mix model-assisted synthesis with human review; see how we verify.
Verification history
We have re-verified FrontierScience 10 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed limitations, switching in, use cases, summary, who should skip it, setup time, our verdict, pricing fit, features, description, hidden costs, pricing tiers, switching out
- — changed pricing page
- — changed our verdict, description, summary, pricing tiers
- — changed pricing tiers, features, description, our verdict, summary
- — changed limitations, pricing page
- — changed who should skip it, description, our verdict, features, who it suits, pricing tiers, summary
Showing the 6 most recent of 10 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does FrontierScience cost in 2026?
FrontierScience: Free to use. Pricing last verified 26 Oct 2026.
Plans compared
Every published FrontierScience tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Free | Free | — | Researchers, biosecurity analysts, and molecular biologists who want to read the full December 2025 evaluation, methodology, and figures without an account or purchaseAdds: Free entry point: an open-access OpenAI research publication with no sign-up, and not a purchasable product or hosted tool |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where FrontierScience's pricing actually pencils out — and where peers do it cheaper.
There is no price to weigh: the work is a free, openly readable OpenAI research publication dated December 16, 2025, not a subscription or per-seat tool. Compare it with what a paid platform would cost you instead — ChatGPT plans sit at the individual level on openai.com/pricing, and lab automation suites are enterprise contracts. FrontierScience's cost shows up as reagents, sequencing, and scientist hours.
Has FrontierScience's pricing changed?
No plan price for FrontierScience has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick FrontierScience?
Pick FrontierScience if you are AI researchers evaluating frontier model reasoning in experimental science, Scientific institutions assessing whether AI can improve real wet lab protocols, Biosecurity analysts working under a Preparedness-style risk framework, Molecular biologists interested in RecA/gp32 homology-based cloning improvements.
Look elsewhere if you are Teams without a wet lab, reagents, and scientists to execute protocols, Researchers who need protocol optimization across many different assays, Anyone expecting a purchasable product rather than a published evaluation, Groups needing results transferable beyond a single GFP/pUC19 cloning setup.
Skip it if Skip FrontierScience if you need to optimize a protocol on your own assay this quarter — the publication gives you the method and the evidence, but no hosted environment, and running the loop yourself means lining up a wet lab and the Red Queen Bio collaboration.
GeneratedReal-world workflow fit
Concrete scenarios for the personas FrontierScience actually fits — and what changes day-one when you adopt it.
You want to test whether a frontier model can improve a real protocol rather than answer biology questions. You read the December 2025 write-up end to end — the 8-10 candidate batches per round, the fixed prompting, the colony-count feedback loop — and adapt the scaffolding for your own evaluation.
Outcome: You get a documented template for an autonomous discovery benchmark plus a concrete result to cite: 79x more sequence-verified clones with n=3 validation error bars.
Your two-piece assemblies are yielding too few verified clones. You read the mechanism section, note that gp32 detangles loose DNA ends before RecA guides each strand to its match, and test the RAPF addition against your current HiFi reaction.
Outcome: A testable hypothesis with published magnitude — 2.6-fold from RAPF assembly and 36-fold from T7 transformation in the GFP/pUC19 system — that you can attempt to reproduce on your own construct.
You are assessing what a frontier model can do in a lab without human steering of its reasoning. You use the scoped, benign design and the Preparedness Framework framing as a reference case in your own review.
Outcome: A worked example of capability evaluation done inside safety constraints, including an explicit statement of what the framework could not do — exploit its own discoveries.
Use cases
- Evaluate a frontier model's ability to propose and iterate on wet-lab protocols
- Assess AI-assisted biology through sequencing-verified rather than narrative results
- Study biosecurity implications of advanced AI in biological research via a scoped, benign system
- Benchmark autonomous discovery loops against static question-and-answer evaluations
- Understand how RecA/gp32 homology-based cloning improves assembly efficiency
- Track progress in AI-assisted experimental design across physics, chemistry, and biology
Setup time & first value
How long it actually takes to get something useful out of FrontierScience — broken out by persona, not the marketing-page minute.
For readers, minutes — the publication is open with no login. For anyone attempting to reproduce the loop, plan weeks: reagents, a functioning HiFi/Gibson cloning workflow, colony-count turnaround on 1-2 day cycles, and sequencing to confirm clones.
Switching to or from FrontierScience
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a static Q&A biology benchmark: reuse the question set as a baseline, then add the execute-and-feed-back step that makes this an online loop rather than a one-shot test
- →From a standard Gibson assembly protocol: start with the published base HiFi reaction, then layer RAPF and T7 as separate variables so the 2.6x and 36x effects can be isolated
- ↗To a closed robotic lab loop: the publication names this as a future direction only, so plan your own automation stack for liquid handling and data upload
- ↗To a general protocol-optimization platform: expect to re-baseline, since the published gains are tied to the GFP/pUC19 cloning setup
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “FrontierScience”, and we withheld 6: 6 could not be judged, because “FrontierScience” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about FrontierScience.
Official links
What are the best FrontierScience alternatives?
The closest alternatives to FrontierScience we list are Triall, Fiddler AI and Goodfire. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| Triall | Freemium | Pricing not published | — |
| Fiddler AI | Freemium | Pricing not published | — |
| Goodfire | Paid | Pricing not published | — |
Alternatives to FrontierScience
View all
Triall
Put a hard question to three frontier models — they answer blind, cross-examine each other, and hand back the one answer that survived, with a verdict.
Fiddler AI
Fiddler AI is an enterprise control plane for agent observability, inline guardrails, and auditable AI governance.
Tools that pair well with FrontierScience
Common stack mates teams adopt alongside FrontierScience, with the specific reason each pairing earns its keep.
Triall
Put a hard question to three frontier models — they answer blind, cross-examine each other, and hand back the one answer that survived, with a verdict.
Fiddler AI
Fiddler AI is an enterprise control plane for agent observability, inline guardrails, and auditable AI governance.
Goodfire
Goodfire's Silico is an AI interpretability agent for reverse-engineering, debugging, and steering model internals.
FrontierScience vs its closest rival
FrontierScience's closest head-to-head is with Surge Ai — If you need to benchmark AI scientific reasoning for free, FrontierScience is the clear choice. But if you're building or aligning frontier AI models and need expert human feedback, RLHF data, or red teaming, Surge AI is far more capable and hands-on—at a premium price. Verdict generated
Head-to-head comparisons
Frontierscience vs Surge Ai
If you need to benchmark AI scientific reasoning for free, FrontierScience is the clear choice. But if you're building or aligning frontier AI models and need expert human feedback, RLHF data, or red teaming, Surge AI is far more capable and hands-on—at a premium price.
GeneratedFrontierscience vs Praktika
These tools serve completely different needs. FrontierScience is a free benchmark for AI scientists evaluating model reasoning, while Praktika is a freemium app for language learners. Your choice depends entirely on whether you are an AI researcher or someone wanting conversational language practice. They are not substitutes.
GeneratedFrequently asked questions
Best-of guides
Topics
Used FrontierScience? Help shape our editorial sentiment research.