Dify pricing

About

Dify is an open-source platform for building LLM applications — chatbots, agents, RAG-backed assistants and multi-step workflows — without writing the orchestration plumbing from scratch. It pairs a visual app studio (Chatbot, Text Generator, Agent, Chatflow, Workflow) with a built-in knowledge base (RAG), model-provider management across OpenAI, Anthropic, Azure OpenAI, Llama 2, Hugging Face and Replicate, and observability that plugs into LangSmith and Langfuse. The same product ships two ways: a managed Cloud Service and a self-hosted edition you run on your own infrastructure.

The company (developer org langgenius on GitHub) positions Dify as a developer-and-team platform that scales from a solo builder on the free Sandbox tier to a 50-person workspace on the Team plan, with Enterprise quoted by sales. Because Dify is open source under the Dify Open Source License (Apache 2.0 with added conditions), it competes both as a commercial SaaS and as a free, self-deployable alternative — and it is resold as “Dify Premium” on the AWS and Azure marketplaces for teams that want a managed self-host with custom branding.

Crucially, Dify’s subscription meters the platform, not the model spend: message credits, apps, seats and knowledge storage are workspace-scoped quotas, while the LLM token costs flow through to whatever model-provider keys the customer brings. That decoupling makes Dify’s own bill predictable even as underlying model usage varies.

Pricing summary : workspace-scoped tiered subscription plus free self-hosted OSS

Dify Cloud is a tiered, workspace-scoped subscription: a free Sandbox tier, Professional at $59 per workspace/month, Team at $159 per workspace/month, and an Enterprise tier via Contact us. Each tier bundles a set of quotas rather than billing per individual seat, so the price is a flat platform fee with capacity caps. Alongside Cloud, the self-hosted Community Edition is free and open source, with paid Premium (marketplace) and Enterprise self-host editions for teams that need branding and governance.

The dimensions that move you between tiers:

• Message credits — 200 one-time (Sandbox) · 5,000/mo (Professional) · 10,000/mo (Team); a workspace meter distinct from the LLM token spend you pay your own model provider.
• Team members (seats) — 1 · 3 · 50; exceeding the cap on a downgrade immediately removes members’ workspace access.
• Apps — 5 · 50 · 200 published applications.
• Knowledge documents — 50 · 500 · 1,000.
• Knowledge data (vector) storage — 50MB · 5GB · 20GB.
• API rate limit — Sandbox is capped at 5,000 calls/month; both paid tiers carry “No Dify API rate limit.”
• Billing cadence — monthly or annual; annual saves ~17% — the compare table shows “Save $118” on Professional and “Save $318” on Team per year.

What makes this different: Dify charges for the app platform and its capacity envelope, not for tokens — a hybrid-versus-pure-subscription choice that keeps the vendor bill flat while customers absorb model costs separately. It is one of the few credit-metered AI platforms whose free open-source self-host sits directly alongside its paid Cloud SKUs.

Pricing by product

Dify Cloud (workspace subscription)

Annual billing saves ~17%: the compare-plans table shows “Save $118 with annual billing” on Professional and “Save $318 with annual billing” on Team. Students, teachers and educational staff can run the Yearly Professional plan free after annual re-verification.

Dify self-hosted (editions)

Sales motions across products: PLG / self-serve for Sandbox, Professional and Team and for the free Community self-host; sales-led for Cloud Enterprise and the self-hosted Enterprise edition; Premium self-host is transacted through the AWS / Azure marketplaces.

Hidden costs : model tokens, the 50-seat Team wall, and DIY self-hosting

The $59 and $159 Cloud prices look like the whole bill, but they only cover the Dify platform — the LLM tokens, the upgrade cliffs, and the operational cost of “free” self-hosting all sit outside the sticker. Two archetypes show where the real spend lands.

Archetype 1 — a small team on Professional that brings its own model keys. Dify meters message credits (5,000/mo on Professional) but the actual model inference is billed by whatever provider key you connect — OpenAI, Anthropic, Azure OpenAI, etc. A team running a RAG chatbot with a frontier model can easily spend more on tokens than on Dify itself:

So the platform fee is the smaller number for any non-trivial workload — Dify’s bill is predictable precisely because the variable, hard-to-forecast cost (tokens) is pushed onto the customer’s own model provider. This is the decoupled-meter pattern that keeps the vendor invoice flat while the real AI cost floats.

Archetype 2 — a 12-person team that hits the Professional seat wall. Professional caps at 3 team members. A team that grows past three has no smooth seat add-on — the docs confirm you can’t independently buy more seats on Professional — so the next step is the Team plan at $159 for up to 50 members:

The jump is mild compared with per-seat SaaS, but it is a hard wall: a 4-person team pays the same $159 as a 50-person team. And on a downgrade, Dify’s docs warn that members exceeding the new plan’s limit lose workspace access immediately — a real operational gotcha when trimming costs.

The “free” self-hosted path removes Dify’s platform fee entirely (Community Edition is OSS), but moves the cost onto your own infrastructure and engineering time: an always-on container, a managed Postgres + vector store, plus patching and upgrades — typically ~$150–400/mo all-in for a single production team, the same band as paid Cloud. Self-hosting wins on raw cost only at higher scale or when data-residency rules require it.

Want to estimate your own Dify bill? Use the Dify pricing calculator to model your monthly cost based on plan tier, message-credit volume, seats and your own model-token spend.

Pricing evolution : stable $59/$159 prices, an evolving meter and a self-hosted split

Dify’s most striking pricing fact is what didn’t change: Professional has been $59/month and Team $159/month since the earliest archived pricing page in December 2023. What evolved instead is the meter (a GPT-call trial became “message credits”), the limit vocabulary (Vector Space became Knowledge Data Storage), and the packaging (a Self-hosted vs Cloud split and a 50-seat Team cap arrived with the Dify 1.0 era in early 2025). The archived raw HTML preserved the tier text and dollar figures legibly across every key snapshot, so the dates below are read from the Wayback record rather than inferred.

Cadence

Tracked range: 2023-12 – 2026-06. Quarters not listed were verified stable (0 price changes, 0 SKU additions). The headline $59 / $159 Cloud prices have not changed across the entire tracked range; annual list prices and the self-hosted Premium price are gated behind Framer toggles that did not render under headless capture and are recorded unknown.

Notable changes

• 2023-04 — langgenius/dify first published on GitHub; the project grows into one of the most-starred AI repos (~143k stars by mid-2026).
• 2023-12-08 — Earliest archived pricing page: Professional $59, Team $159, Sandbox free — the prices Dify still charges today. (Wayback raw HTML.)
• 2024-04-22 — A 185-point Hacker News thread (“Dify, a visual workflow to build/test LLM applications”) puts Dify in front of the developer mainstream.
• 2024-06 — Value metric relabelled to “Message Credits” (200 / 5,000 / 10,000). (Wayback raw HTML.)
• 2025-03 — Dify 1.0 ships (plugins, marketplace, new agent node, optimized RAG/memory); the pricing page is rebuilt with a Self-hosted/Cloud toggle, per-workspace framing, Knowledge Data Storage limits and a 50-seat Team cap. (Wayback raw HTML; Dify changelog.)
• 2026-06-03 — Current capture: same $59 / $159 prices; education program gives the Yearly Professional plan free to verified students/educators.

The license-and-self-host model in detail

Dify ships under the Dify Open Source License — Apache 2.0 plus two added conditions read directly from the repo LICENSE: (1) you may not operate Dify as a multi-tenant / managed SaaS without written authorization, and (2) you may not remove or modify the Dify logo or copyright in the console or web app. Everything else inherits Apache 2.0’s permissiveness. That single restriction is the commercial moat: anyone can self-host the Community Edition for free, but reselling it as a hosted service is off-limits, which protects Dify Cloud and the marketplace-billed Dify Premium (AWS/Azure) without resorting to a closed core. It is the same open-core logic n8n and other infra-OSS vendors use, expressed as a source-available license rather than a feature-gated free tier.

What’s unique : platform-metered, token-decoupled, OSS-on-the-same-page pricing

1. The subscription meters the platform, not the model spend. Dify charges for message credits, apps, seats and storage — workspace quotas — while the LLM tokens flow through to the customer’s own model-provider key. That decoupling of value metric from cost driver keeps Dify’s invoice flat and predictable even as a customer’s underlying model usage swings, the opposite of token-pass-through platforms whose bill tracks every inference.

2. Headline prices that have not moved in 2.5+ years. Professional ($59) and Team ($159) have held since the December 2023 archive while the product added plugins, a marketplace, agent nodes and a self-hosted split. Price stability through heavy feature growth is itself a positioning choice: Dify competes on capability and OSS goodwill, not on repricing.

3. Free open-source self-host sitting on the same pricing page as paid Cloud. The Self-hosted vs Cloud Service toggle puts a $0 OSS option one click from the $59/$159 cards. Dify openly competes against its own free edition, monetizing managed hosting, governance and marketplace branding rather than core capability — a freemium/open-core bet that the free tier feeds the funnel instead of cannibalizing it.

4. A “message credit” that abstracts away token complexity. Rather than expose raw tokens, Dify counts message credits (200 / 5,000 / 10,000) as a coarse, human-legible unit — a deliberately simpler credit-based meter that lets a buyer reason about “thousands of messages” instead of millions of tokens, while the true token cost is handled separately by their provider.

5. Giving the paid mid-tier away to an entire buyer segment. Verified students, teachers and educational staff run the Yearly Professional plan free (re-verified annually). Handing a usage-metered platform’s mid-tier to a whole segment is a long-game adoption play aimed at the developers who will later specify tooling at work.

Strengths & weaknesses

Billing UX : workspace-scoped subscription controls in Settings > Billing

Billing in Dify Cloud is workspace-scoped — the subscription governs the whole workspace’s member limit, feature access and usage quotas. Named controls observed on the pricing page and the subscription-management docs:

• Cloud Service vs Self-hosted toggle — top-of-page pill that switches the displayed plan set between the managed Cloud tiers and the self-hosted editions.
• “Bill Annually — Save 17%” switch — flips Professional and Team between monthly and annual pricing (annual saves $118 and $318 respectively).
• Settings > Billing — Upgrade — only workspace owners and admins can access billing settings or change plans; upgrades take effect immediately and are prorated for the current period.
• Plan-change handling — downgrades take effect immediately with billing adjustments on the next invoice; members exceeding the new plan’s seat limit lose workspace access immediately.
• Cancellation behavior — a cancelled paid plan continues until the end of the current cycle, then reverts to the free Sandbox plan.
• Education discount control — Settings > Billing > “Get Education Verified” applies the free Yearly Professional plan to a chosen workspace after school-email verification.
• Invoices & receipts — surfaced in the billing settings (the pricing FAQ addresses where to find invoices and which payment methods are accepted).
• Message-credit meter — per-workspace usage counter against the tier’s monthly credit allotment, separate from the customer’s own model-provider token spend.

Strategic wins : why the decoupled meter and OSS funnel worked

1. Decoupling the platform fee from the model spend

By metering message credits, apps, seats and storage — and pushing LLM tokens onto the customer’s own provider key — Dify keeps its invoice flat and forecastable while the volatile cost floats elsewhere. That is the right value-metric instinct: charge for the thing the buyer can predict and control, not the thing that spikes with every inference. It also sidesteps the bill-shock problem that plagues token-pass-through platforms.

2. Putting a free OSS self-host on the same page as paid Cloud

Dify’s Self-hosted/Cloud toggle openly offers a $0 open-source path next to the $59/$159 cards, betting that the free edition feeds the funnel rather than draining it. The paid story is operational convenience — managed hosting, marketplace branding, governance — which is the open-core variant of usage-based packaging that converts free self-hosters into Cloud customers once running infrastructure stops being worth the trouble.

3. Holding headline prices steady through heavy feature growth

Professional and Team have stayed $59 and $159 since 2023 while Dify shipped plugins, a marketplace, agent nodes and a self-hosted split. Stable prices through capability growth build buyer trust and remove a recurring objection — the inverse of vendors who reprice with every release. It signals that Dify intends to win on product and community, an approach worth comparing across the pricing blueprint.

4. Turning the license into the moat instead of a closed core

The Dify Open Source License (Apache 2.0 + no-multi-tenant-resale + no-logo-removal) lets the whole product stay source-available while still protecting Cloud and the marketplace-billed Premium edition. Dify monetizes hosting and branding, not gated features — a cleaner open-core posture than feature-crippling a free tier, and a big reason the repo earned ~143k GitHub stars.

Areas to improve : seat-cliff, price transparency and credit-to-cost mapping

1. Add a seat add-on between Professional and Team

Professional caps at 3 members with no independent seat purchase, so a 4-person team must jump to the $159 Team plan. A per-seat add-on or a small mid-tier would catch teams of 4–10 who currently overpay for 50-seat headroom they don’t need. Pairing that with thresholding and alerting as a workspace approaches its seat or credit limit would soften the wall.

2. Publish annual and self-hosted Premium prices on the page

Annual list prices and the AWS/Azure-billed Premium price are hidden behind Framer toggles and marketplace listings — a buyer can’t see the real annual number or the Premium cost without leaving the page. Surfacing them inline (the compare table already shows the $118/$318 annual savings) would cut friction the way the better usage-invoicing playbooks recommend.

3. Make the credit-to-token-cost relationship explicit

Because tokens are billed by the customer’s own provider, a buyer can’t easily reason about what 5,000 message credits will actually cost them in model spend. A simple “a message credit is roughly N tokens / $X at model Y” estimator on the pricing page — or in the calculator — would close the gap between the legible message-credit meter and the real, off-bill AI cost.

Key takeaways

1. Decouple the meter from the cost driver. Dify charges for platform capacity and lets model tokens flow through to the customer’s own provider — keeping the vendor invoice predictable while the volatile cost floats. The “right” unit is the one the buyer can forecast.
2. Price stability is a feature. Holding $59/$159 for 2.5+ years through major feature releases builds trust and removes a recurring buyer objection; not every release needs a reprice.
3. A free OSS tier can be a funnel, not a leak — if the paid value is operational. Dify sells managed hosting, marketplace branding and governance, not gated capability, so the free self-host feeds Cloud instead of cannibalizing it.
4. A coarse, human-legible unit beats raw tokens for buyer comprehension. “Message credits” let a buyer reason in thousands of messages rather than millions of tokens — but the abstraction has to be mappable back to real cost or it just hides the bill.
5. Watch the seat cliffs your tiers create. A clean two-tier Cloud lineup still forces a 4th user from $59 to $159; legible axes don’t guarantee legible steps, and the gaps are where friction concentrates.

UBP implications

1. Platform-metered + token-pass-through is a distinct AI pricing pattern. Charging for the orchestration layer while the model spend is the customer’s own gives predictable vendor revenue and predictable buyer platform cost — at the price of the sticker understating true spend.
2. Source-available licensing can substitute for a feature-gated free tier. When the moat is “you can’t resell this as SaaS,” a vendor can keep the whole product open and still protect its managed offering — shifting the pricing job from capability-gating to operational convenience.
3. Credit abstraction trades transparency for comprehension. Coarse credits make usage legible but obscure the underlying cost; UBP designers must decide how much real-cost visibility to give up for a friendlier unit.

Sources

• Dify pricing page (accessed 2026-06-03)
• Dify subscription & billing docs (accessed 2026-06-03)
• Dify changelog (accessed 2026-06-03)
• Dify blog (accessed 2026-06-03)
• Dify Open Source License (langgenius/dify LICENSE) (accessed 2026-06-03)
• Dify on GitHub (langgenius/dify) (accessed 2026-06-03)

Bottom line

Dify prices the LLM-app platform itself — message credits, apps, seats and storage on a flat $59/$159 Cloud subscription — while the real model-token cost flows through to the customer’s own provider, and a free open-source self-host sits one click away on the same page. Headline prices that haven’t moved since 2023, a source-available license as the moat, and ~143k GitHub stars make it one of the cleanest examples of monetizing an AI platform’s convenience rather than its capability.

Want to compare Dify against other LLM-app and automation pricing? Browse the pricing blueprint.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Current Cloud pricing snapshot — Sandbox / Professional $59 / Team $159

Jun 2026

Dify Cloud offers a free Sandbox tier, Professional at $59 per workspace/month and Team at $159 per workspace/month, with Enterprise via Contact us. Tiers are metered on message credits, apps, team members, knowledge documents and knowledge data storage; annual billing saves ~17% (Professional 'Save $118', Team 'Save $318' per year). Self-hosted Community Edition remains free and open source. The annual list prices and the self-hosted Premium card price are Framer toggles that did not render under headless capture and are recorded unknown.

Workspace-scoped repackaging + Self-hosted toggle (around Dify 1.0)

Mar 2025

Between the 2025-01-12 (old layout) and 2025-03-20 snapshots Dify rebuilt the pricing page: a Self-hosted vs Cloud Service toggle appeared, prices moved to 'per workspace/month', storage was relabelled 'Knowledge Data Storage' (50MB / 5GB / 20GB), Knowledge Documents (50 / 500 / 1,000) replaced the old upload quota, and the Team plan's seat ceiling was capped at 50 (down from Unlimited). The $59 / $159 prices did not change. This coincides with the Dify 1.0 release (plugins, marketplace, new agent node) around 2025-03-01. (Wayback raw HTML, 2025-03-20.)

Meter renamed to 'Message Credits' (200 / 5,000 / 10,000)

Jun 2024

By the 2024-06-01 snapshot the value metric was relabelled from a GPT-call trial to 'Message Credits' — Sandbox 200 messages, Professional 5,000/month, Team 10,000/month — the structure Dify still uses. Prices held at $59 and $159. Team Members were still 'Unlimited' on the Team plan and storage was still labelled 'Vector Storage'; no Self-hosted toggle existed yet. (Wayback raw HTML, 2024-06-01.)

Earliest archived pricing — Sandbox / Professional $59 / Team $159 (Vector Space meter)

Dec 2023

The earliest Wayback snapshot of dify.ai/pricing (2023-12-08) already shows Professional at $59/month and Team at $159/month — the same headline prices Dify charges today. Sandbox was a free '200 times GPT free trial', and tiers were differentiated by Vector Space (10MB / 200MB / 1GB), Build Apps (10 / 50 / Unlimited) and Team Members (1 / 3 / Unlimited). Yearly billing gave '2 months free' (~17% off). (Wayback raw HTML, 2023-12-08.)