Offering

Every product side by side, with access and indicative cost.

Standard AI defaults to the median. Novamine’s architecture is engineered to break that consensus. The grid below maps the escalation of intelligence, from baseline models on the left to the full Novamine enterprise engagement on the right. Compare the engines, review the independent NoveltyBench scores, and select the architecture built for your domain.

The Products side by side, with access and indicative cost

Every product in one grid, from standard AI on the left to the full Novamine engagement on the right. Geode, Novacore and Novacrowd are the three API products; all are built on the Novamine engine that took No.1 on NoveltyBench. Read down for the benchmark scores, run time, fact checking, the industries each suits, then how to get it and what it costs.

Feature	Standard AI Any frontier model, asked directly. The benchmark baseline. The field	Geode Creative ideation for agencies. Three novelty levels, powered by Novamine. API	Novacrowd Pools of distinct, scored answers for your own aggregator, MoA or MAD stack, judge or forecasting bot. Diversity as a supply. API	Novacore Serious science, finance, strategy and law. Three novelty levels, four levels of hardening. API	Novamine Full human-in-the-loop R&D consultancy engagement Engagement
		Built on Novamine APIs, holder of the current Carnegie Mellon NoveltyBench No.1 for novelty and utility: the same Novamine engine under all three products

Output	What each offering hands back
What You Get	One response, three paragraphs of consensus The statistically safest reading Ask ten times and you get the same idea in different words	One distinct idea per response; one to six responses a run Titled and argued in a few paragraphs each Four novelty levels: Refract, Fluoresce, Scintillate, Fever Dream Each scored for novelty and utility Optional claim count and brand-name clash check No rewriting of facts	A pool, not a single answer: up to 60 concepts or up to 48 developed answers per brief Every item distinct by construction Each scored for novelty and utility, or unscored for your own judge Five speeds: Light and Concepts (unscored), Standard (scored), Developed, Hardened Score and harden routes for your own answers Light and Concepts carry no scores: at that speed there is no judge pass, and Light runs on a lighter model, so novelty and utility are yours to judge No synthesis: your aggregator does that	One developed response, hardened; one to six a run Titled and argued over several paragraphs Recency checked, facts checked against live sources, hardened Each scored for novelty and utility Three novelty levels: Grounded, Adventurous, Unprecedented Four levels of hardening: Brainstorm only, Standard, Deep, Maximum	Up to 20 verifiable novel theses per engagement Free of precedent and prior art: IP-ready Each with its mechanism, its decisive experiment and a testable validation roadmap Every claim red-teamed by the adversarial network, with our people in the loop Under strict NDA

Scores	NoveltyBench judge v1.1, out of 10. Standard AI is the incumbent frontier model at its best on the board; asked raw it scores 2.4 and 2.9. Novamine scoring where no official run exists.
Typical NoveltyDistinct score	6.68NoveltyBench, incumbent frontier model	8.5the number Geode is built for; Levels 1 and 2; 8.0 at Level 3	8.0the pool average, Developed; single items range 7 to 9. Light and Concepts unscored	8.5Unprecedented; about 7.8 at Grounded	9.5+Novamine scoring
Typical UtilityUsefulness score	5.32NoveltyBench, incumbent frontier model	7.5Levels 1 and 2; 7.0 at Level 3	7.0Developed; 8.0 and above Hardened	8.0the number Novacore is built for; Standard hardening and above	9.5Novamine scoring

Speed and Checks	How long a run takes, and whether the responses are checked against sources.
Typical Run Time	Under a minute	3 to 5 minutes	About 2 minutes Light, 3 minutes Concepts, 7 Standard 5 to 15 minutes Developed, 20 to 60 Hardened	A few minutes at Brainstorm only About 20 at Standard, up to an hour at Maximum	24 hours and more
Fact CheckingResponses checked against sources	No	Optional claim count and brand-name clash, no rewriting	Hardened only live fact check and hostile review at the hardened speeds	Yes recency, live fact check, hardening	Full adversarial An agentic red team attacks every claim until only the defended ones survive; our people in the loop

Industry Presets	Which industries each offering suits. Gold tick: yes. Grey cross: no.
ScienceR&D	No	No	Yes	Yes	Yes
Strategy	No	No	Yes	Yes	Yes
Finance	No	No	Yes	Yes	Yes
Legaland IP	No	No	Yes	Yes	Yes
AdvertisingMarketing, creative	No	Yes	No	No	No
Suitable For	Basic and deep research of precedent Novelty: the same first response for everyone who asks, low by design	Agencies and in-house creative teams Campaign territories, platforms, names and design directions Pick how far from the obvious: Refract, Fluoresce, Scintillate or Fever Dream Science, finance or legal work	Mixture-of-Agents and multi-agent stacks that need proposers which disagree Forecasting bots and platforms: rationale pools per question Research systems that rank and debate hypotheses Analyst teams wanting the options they would not have listed Anyone wanting one finished answer: that is Novacore	Science, legal, finance and strategy briefs that will be challenged Further from the obvious, facts checked, hardened Harder testing for the brief that will be challenged Advertising briefs, which Geode serves	Science: mechanism-level hypotheses, each with its decisive experiment, defended by the adversarial network Legal and IP: positions and design-arounds that survive challenge Finance: theses with every number traced Strategy: moves with the counter-moves already tested Creative agencies: the depth is wasted on a campaign brief

Deploying In Your Organisation	How to get each offering, the shape it takes in your stack, and the API setting to send.
How to Access It	Available to the public from any AI provider	Request a key for the demo Submit your query Request access to the full API or chat interface	Request a client key Send a brief, receive the pool Ask for a pilot on your own evaluation	Request a key for the demo Submit your query Request access to the full API or chat interface	Sign the NDA Call with us Agree IP retention First run: iterative Second or third run: final exchange
Engagement Type	Any chat product	API or white label page	API, batch jobs, or a pilot on your evaluation set	API or white label page	Human-in-the-loop consultancy
API SettingThe product and level to send	n/a	product=geode level=refract \| fluoresce \| scintillate \| feverdream	mode=light \| concepts \| standard \| developed \| hardened domain=science \| strategy n=1 to 60	product=novacore domain=science \| strategy novelty=grounded \| adventurous \| unprecedented hardening=brainstorm \| standard \| deep \| maximum	n/a

Commercial Tiers	Price per response, all in, by monthly plan
Tier A£0 ($0) / mo, pay as you go	n/a	£1.40 - all novelty levels, 1 response ($1.90)	£0.40 - Light, per pool, unscored ($0.55) £1.00 - Concepts, per pool, unscored ($1.35) £2.80 - Standard, per pool, scored ($3.80) £3.50 - Developed, per answer ($4.75) £6.70 - Hardened, per answer ($9.05)	£5.50 - Brainstorm only, 1 response ($7.45) £11.00 - Standard ($14.85) £16.50 - Deep ($22.30) £20.50 - Maximum ($27.70)	From £10,000($13,500) package, dependent on IP retentionTalk to us
Tier B£95 ($129) / mo	n/a	£1.20 - all novelty levels, 1 response ($1.60)	£0.35 - Light, per pool, unscored ($0.45) £0.85 - Concepts, per pool, unscored ($1.15) £2.50 - Standard, per pool, scored ($3.40) £3.15 - Developed, per answer ($4.25) £6.00 - Hardened, per answer ($8.10)	£5.00 - Brainstorm only, 1 response ($6.75) £10.00 - Standard ($13.50) £15.00 - Deep ($20.25) £18.50 - Maximum ($25.00)
Tier C£295 ($399) / mo	n/a	£1.00 - all novelty levels, 1 response ($1.35)	£0.30 - Light, per pool, unscored ($0.40) £0.70 - Concepts, per pool, unscored ($0.95) £2.10 - Standard, per pool, scored ($2.85) £2.60 - Developed, per answer ($3.50) £5.00 - Hardened, per answer ($6.75)	£4.20 - Brainstorm only, 1 response ($5.65) £8.30 - Standard ($11.20) £12.50 - Deep ($16.90) £15.30 - Maximum ($20.65)
Unlimitedper seat, per month	n/a	£2,000fair use 400 runs a month	£3,500fair use 500 pools a month at Light, Concepts and Standard; Developed and Hardened metered	£4,500fair use 150 runs a month, any level	n/a

Deploying This in Your Organisation an API in your stack, or hosted by us

Geode and Novacore ship in two shapes. The simplest is a hosted page like this one, re-voiced and branded for your organisation, with access codes for your team: nothing to install, working the day you sign. The deeper lane is a REST API your own tools call, which your engineers deploy however suits: an intranet page, a Slack bot, a tool your existing AI assistant invokes mid-conversation, a batch pipeline. If your systems team wants the engine wired into something specific, that is scoped integration work we take on with them. The full Novamine engagement is not an API: it starts with an NDA and a call, as the table above sets out.

For your engineers: one endpoint, POST /v1/generate, with simple request parameters: product (geode or novacore), level (refract, fluoresce, scintillate or feverdream for Geode), domain (science, or strategy for strategy, finance and law; read off the brief if omitted), novelty (grounded, adventurous or unprecedented for Novacore), hardening (brainstorm, standard, deep or maximum for Novacore), k (responses per run: 2 by default, up to 6) and, for Geode, fact_check (true for the claim count and name-clash check). Novamine chooses and runs the engine; you never manage a model. Input and output length can be capped per call or per deployment for token control. An OpenAI-compatible endpoint lets it drop into existing tooling unchanged, and it speaks standard tool-calling, so the AI assistants your organisation already runs can invoke it mid-conversation. Long runs are available as asynchronous jobs with status polling, and per-key allowances and full usage logs come as standard.

Novamine's fee is the flat per-generation amount in the tier columns of the table above, plus the monthly tier; a higher tier buys the per-generation fee down, and the same rate applies hosted and API alike. Tier A is capped at 100 generations, enough to prove it on live work, after which we ask you to choose a tier.
Every figure is indicative, negotiable and quoted per organisation, with heavy client-specific tailoring: deployment shape, volume commitments, engine mix, input and output token caps and support all move the numbers. Nothing here is a published price list.

Novamine, the full engagement →Try Novacore →Try Geode →The science behind it →