Inflection AI pricing

About

Inflection AI is a US foundation-model lab whose story is defined by an unusual corporate fracture. Founded in 2022 by Mustafa Suleyman (ex-DeepMind), Karén Simonyan, and Reid Hoffman, it built Pi — a free, deliberately empathetic personal-AI assistant at pi.ai — on top of its own Inflection-1 and Inflection-2.5 models. In 2023 it raised $1.3B from Microsoft, Nvidia, Bill Gates, and Eric Schmidt at a reported ~$4B valuation, making it one of the best-capitalized consumer-AI bets of the era. It had no priced product: Pi was free and the company ran on venture capital.

That changed abruptly in March 2024, when Microsoft executed what is widely described as an “acqui-hire without an acquisition.” Microsoft hired Suleyman, Simonyan, and most of the ~70-person staff into a new Microsoft AI division (Suleyman became its CEO), and paid Inflection roughly $650M — $620M for a non-exclusive technology license plus $30M to agree not to sue over the poaching. Inflection remained a legally separate, independent company, but lost its founding team and its consumer momentum overnight. Investors were made roughly whole rather than rewarded (early-round ~1.5x, $1.3B-round ~1.1x).

Under new CEO Sean White, Inflection pivoted hard from consumer to enterprise. In October 2024 it launched Inflection 3.0 and Inflection for Enterprise: a two-model family (Pi 3.0 for empathetic support, Productivity 3.0 for structured, instruction-following output) deployable on Intel Gaudi 3 hardware on-premises, in the cloud, or hybrid. The pitch inverts the cloud-rental norm — enterprises own the software and the appliance for lower total cost of ownership rather than paying per token to a hyperscaler. The consumer Pi app still exists but is de-prioritized, and Inflection no longer publishes a pricing page at all.

Pricing summary : from free consumer Pi to sales-only enterprise

Inflection AI’s current pricing model is sales-only: there is no public pricing page (inflection.ai/pricing returns a 404), and the enterprise product is quoted entirely through a “Get in touch” sales contact. The dimensions, as far as they are observable, are:

• Owned / on-prem deployment — Inflection for Enterprise sells the Inflection 3.0 models as software you own and run on a purpose-built Intel Gaudi 3 appliance, on-premises, in the cloud, or hybrid. The value metric is total cost of ownership versus renting cloud inference — a capital-and-commitment model, not a per-seat or pay-as-you-go one.
• Custom fine-tuning — models are “fine-tuned to your business,” grounded in your data and aligned to internal policies, sold as part of a bespoke engagement rather than a self-serve product.
• Legacy per-token API (withdrawn) — at the October 2024 launch, the API was briefly public at $2.50 per 1M input tokens and $10 per 1M output tokens for both Inflection 3.0 models (8k context). That public card has since been removed.
• Pi consumer — free, with no paid tier; not a monetized surface.

What makes this different: Inflection is the rare frontier-model lab whose pricing story is a retreat from transparency — it had a public per-token card and a free consumer hit, then withdrew both in favor of a quoted, owned-deployment enterprise model after the Microsoft acqui-hire gutted its consumer ambitions.

Pricing by product

Inflection for Enterprise (current — sales-only)

Sales motions across products: fully sales-led for Inflection for Enterprise — there is no self-serve path into the core enterprise technology, so every engagement runs through a sales process and a custom quote. The free Pi consumer app is the only non-sales surface, and it is not monetized.

Legacy Inflection 3.0 API (October 2024 launch — public rate, since withdrawn)

These were the public per-million-token rates at launch; both models shared the same rate. The public card has since been removed and is shown here for historical reference only.

Pi consumer assistant

Hidden costs : What Inflection AI customers actually pay

With no public price card, the “hidden cost” question for Inflection is really a question about the shape of an owned-deployment deal versus the cloud-rental alternative the company is pitching against. Two reference points illustrate the trade.

Reference 1 — what the legacy API would have cost. Before the public card was withdrawn, a workload of ~60M input + ~15M output tokens a month on Inflection 3.0 would have billed as below. It is a useful anchor because the enterprise quote is implicitly benchmarked against renting this kind of capacity.

The lesson in the rate itself: at $10/M, the output rate is 4x the input rate, so output-heavy workloads dominate — the same premium structure as most frontier APIs, despite a modest 8k context window.

Reference 2 — the owned-appliance TCO bet. Inflection for Enterprise asks buyers to convert that recurring per-token bill into a capital-and-commitment purchase: an Intel Gaudi 3 appliance you own, deployed on-prem for data locality. The “hidden” costs there are the ones every owned-infrastructure deal carries — hardware lifecycle, ops staffing, and utilization risk — traded against the promise of lower long-run TCO and no per-token meter. Because the deal is fully quoted, the real cost is whatever the sales negotiation lands on; there is no published floor to anchor against.

Want to estimate your own Inflection AI bill? Use the Inflection AI pricing calculator to model the legacy per-token rate against an owned-deployment scenario.

Pricing evolution : Inflection AI pricing history and changes

Inflection’s pricing history is a clean arc through three regimes: a free, venture-funded consumer phase (Pi, 2023); a brief public per-token enterprise API at the Inflection 3.0 launch (October 2024); and a withdrawal into fully sales-only custom quoting (2025 onward). The Microsoft acqui-hire in March 2024 is the hinge between phases.

Cadence

Tracked range: 2023 Q2–2026 Q2. Quarters not listed had no publicly announced price or SKU change. The 2025-Q2 “price change” is the removal of the public card, not a rate change.

Notable changes

• 2023-05 — Pi launches free on Inflection-1; the company is consumer-first and venture-funded, with no priced product.
• 2023-06 — $1.3B raised from Microsoft, Nvidia, Bill Gates, and Eric Schmidt at a reported ~$4B valuation.
• 2024-03-21 — Microsoft acqui-hire: ~$650M ($620M license + $30M non-litigation); Suleyman and Simonyan plus most staff move to Microsoft AI; Sean White becomes CEO (reported by Bloomberg and TechCrunch).
• 2024-10-07 — Inflection 3.0 and Inflection for Enterprise launch with a public API at $2.50 in / $10 out per 1M tokens, on Intel Gaudi 3 (reported by Maginative and The Register).
• 2025-01 — Gaudi 3 on-prem AI appliance ships, anchoring the “own it, don’t rent it” TCO pitch.
• 2025-06 — Public pricing withdrawn; the only way to buy becomes a sales quote, and price_transparency flips to sales-only (per eesel.ai’s pricing guide).

The Microsoft acqui-hire → enterprise pivot in detail

The March 2024 deal is the most consequential pricing event in Inflection’s life precisely because it destroyed the consumer pricing question. Microsoft licensed the technology non-exclusively for $620M and hired away the team that had built Pi, paying a further $30M to keep the peace — a structure designed to deliver “a good outcome” for VCs (per Reid Hoffman) without triggering an antitrust-sensitive acquisition. With its consumer flywheel and founders gone, Inflection had no path to monetize Pi at scale. Sean White’s pivot to owned, on-prem enterprise deployments is a direct response: rather than compete for consumer subscriptions against ChatGPT and Pi’s former self, Inflection sells fine-tuned, data-resident models that large regulated buyers can own outright. The withdrawal of the public per-token card a year later completes the move — from a transparent, self-serve consumer posture to an opaque, quote-driven enterprise one.

What’s unique : Inflection AI’s distinctive pricing mechanics

1. A retreat from price transparency, not toward it. Almost every lab in this cluster moves toward published rates as it matures. Inflection went the other way: it had a free consumer hit and a public per-token API, then withdrew both for a sales-only enterprise quote. The acqui-hire severed the consumer business from the people who could grow it, and the pricing model followed — opacity as a consequence of corporate fracture rather than a deliberate land-grab.

2. “Own it, don’t rent it” TCO as the value metric. Instead of metering per token, Inflection for Enterprise sells an owned Intel Gaudi 3 appliance pitched on lower total cost of ownership than cloud rental. The value metric is effectively avoided per-token spend plus data locality — a capital-purchase framing that’s unusual for a frontier-model lab and aligns the vendor with on-prem, sovereignty-sensitive buyers rather than usage-growth.

3. A withdrawn public anchor that still benchmarks the quote. The defunct $2.50-in / $10-out per-1M-token card hasn’t vanished from buyer memory. It functions as a shadow anchor: enterprise quotes are implicitly compared against what renting that capacity would cost, even though Inflection no longer lists it. The per-token economics became invisible on the page but remain the mental yardstick.

Strengths & weaknesses

Billing UX : usage tracking and procurement

• No self-serve billing surface — there is no public pricing page, no sign-up-and-pay path into the enterprise product; procurement runs entirely through a “Get in touch” sales contact.
• Owned-appliance model — billing for the core offering is structured as a capital/commitment purchase of an Intel Gaudi 3 appliance plus software, not a metered monthly invoice.
• Custom contracts — terms, support, and fine-tuning scope are negotiated per deal; expect enterprise procurement, MSAs, and committed-use structures rather than usage dashboards.
• Legacy API metering — the withdrawn public API billed per million input/output tokens ($2.50 / $10), the standard token-meter UX, before the card was removed.
• Pi consumer — free, no billing; not a paid surface.

Strategic wins : Why Inflection AI’s pricing decisions worked

1. Turning a gutted consumer business into a defensible enterprise niche

Stripped of its founders and consumer flywheel, Inflection could have folded. Instead it repriced around what it still had — proprietary models and Microsoft/Nvidia/Intel relationships — and sold owned deployments to buyers cloud-native rivals underserve. Repositioning the value metric from consumer engagement to enterprise TCO was the move that kept the company alive. See usage-based pricing strategy for why matching the meter to the surviving asset matters.

2. “Own it, don’t rent it” as a wedge against hyperscalers

By selling an on-prem Gaudi 3 appliance pitched on lower TCO than cloud rental, Inflection carved out a position the per-token cloud labs structurally can’t occupy. For regulated buyers with steady, high-volume inference and data-locality needs, a capital purchase can genuinely beat a usage meter — a deliberate counter to the per-seat and per-token consensus.

3. Letting the deal define the price

For a small lab selling to large regulated accounts, fully custom quoting captures more value than a published rate would — each deployment is scoped to the customer’s data, hardware, and compliance needs. Going sales-only is a defensible choice when the buyer is an enterprise that expects a negotiated contract anyway. Related: choosing the right usage metric.

Areas to improve : Gaps in Inflection AI’s pricing approach

1. Restore a public anchor or worked example

With no pricing page, mid-market buyers can’t self-qualify and many won’t book a call. Republishing the per-token rate, or a worked owned-deployment TCO example, would shorten the evaluation cycle and reduce exactly the bill-shock and unpredictability anxiety that quote-only models create.

2. Resolve the Pi consumer ambiguity

Leaving a free, de-prioritized consumer app live with an unclear roadmap is a brand liability — it neither monetizes nor signals direction. Either sunsetting Pi cleanly or giving it a defined (even free) role in the enterprise funnel would remove strategic ambiguity.

3. Publish the context and capability story behind the quote

Buyers comparing Inflection 3.0’s 8k context against frontier peers need the deployment, fine-tuning, and TCO benefits made concrete to justify the trade. A transparent capability-and-cost narrative would help the owned-model pitch stand on its own rather than relying on a sales conversation to surface it. Compare how other AI companies stage enterprise transparency.

Key takeaways

1. A pivot can run pricing backwards. Inflection went from a free consumer hit and a public per-token API to a sales-only quote — proof that maturity doesn’t always mean more transparency; corporate fracture can force the opposite.
2. Match the meter to the surviving asset. After losing its consumer flywheel, Inflection repriced around owned models and hardware partnerships — the assets it still controlled — rather than chasing the consumer market it had lost.
3. “Own vs rent” is a real pricing axis. For steady, high-volume, data-resident workloads, a capital-purchase appliance can beat a per-token meter — a counter-position to the cloud-rental default worth watching.
4. A withdrawn price still anchors. Removing the $2.50/$10 card from the page didn’t remove it from buyers’ heads; quotes are still mentally benchmarked against the rental alternative.
5. Trust events are pricing events. Microsoft’s $650M acqui-hire didn’t just move people — it destroyed the consumer pricing question and dictated the enterprise model that replaced it.

UBP implications

1. Usage-based pricing presumes a usage business — lose that, lose the meter. Inflection’s retreat from per-token billing shows that when the consumer/usage flywheel disappears, the natural model shifts toward owned deployments and custom contracts. UBP only fits where there’s durable, self-serve volume to meter.
2. Owned-deployment TCO is the anti-UBP for some buyers. For regulated, high-volume, data-resident workloads, converting a recurring per-token bill into a capital purchase can be the rational choice — UBP strategists should know when their buyer would rather own than rent.
3. A public rate is an asset even after you remove it. The withdrawn $2.50/$10 card still functions as a benchmark. The lesson for UBP design: published rates create durable anchors, so withdrawing them sacrifices a comparison tool you don’t fully get back.

Sources

• Inflection for Enterprise (accessed 2026-06-11)
• Inflection AI homepage (accessed 2026-06-11)
• Maginative — “Inflection AI Unveils Enterprise Offering with New Inflection 3.0 Models” (accessed 2026-06-11)
• The Register — “Inflection AI’s new offering ditches Nvidia for Intel Gaudi” (accessed 2026-06-11)
• Bloomberg / Yahoo — “Microsoft to Pay Inflection AI $650 Million After Scooping Up Most of Staff” (accessed 2026-06-11)
• TechCrunch — “Here’s how Microsoft is providing a ‘good outcome’ for Inflection AI VCs” (accessed 2026-06-11)
• eesel AI — “A guide to Inflection AI pricing in 2025” (accessed 2026-06-11)
• Browse the pricing blueprint corpus

Bottom line

Inflection AI is the cluster’s cautionary tale: a free consumer hit (Pi) and a $4B valuation, then a March 2024 Microsoft acqui-hire that took its founders and most of its staff for ~$650M and left the company to pivot. Under Sean White it now sells sales-only, owned/on-prem Inflection 3.0 deployments on Intel Gaudi 3, pitched on lower TCO than cloud rental — and it no longer publishes a price at all. The legacy $2.50-in / $10-out per-1M-token API is the only public anchor it ever had, and it’s been withdrawn. The pricing story here is a retreat from transparency forced by a corporate fracture, not a strategy chosen from strength.

Want to compare Inflection AI against other foundation-model providers? See Mistral AI and OpenAI, or browse the full pricing blueprint.