AI Pricing Cheat Sheet: Compare ChatGPT, Claude, Gemini, and More

If you've tried to figure out what an AI subscription actually costs this year, congratulations, you've just discovered the internet's newest blood sport.

Sixteen months ago, "getting AI" meant picking between ChatGPT Plus and maybe Midjourney. Today, it means choosing among chat assistants, image generators, and video tools that span at least three pricing philosophies: flat monthly fees, opaque credit systems, and per-token API billing, built by companies headquartered in places ranging from San Francisco to Paris to Hangzhou. 

Prices have also started moving fast. Several platforms have changed their entire tier structures in the past few months alone, and at least two companies — OpenAI and Google — have recently split their top consumer tier into two separate price points rather than one.

That matters because the gap between the cheapest usable plan and the most expensive one is no longer small. A casual user can get real work done for $0 to $20 a month. A power user chasing the newest video or reasoning models can easily clear $200 a month on a single tool, and several hundred dollars if they're running three or four platforms at once. 

Multiply that across a household, a marketing team, or a small agency, and the AI tooling line item starts to rival what companies used to spend on an entire software stack.

How I built this cheat sheet

I checked every price below directly against each company's own pricing page, help center documentation, or developer docs, rather than pulling it from a roundup or a press release. Where a platform uses a confusing or rapidly changing structure, and several do, I've noted that explicitly. 

AI pricing in 2026 changes often enough that a number that's accurate today could be stale in a month, so treat this as a snapshot rather than gospel, and double-check the linked page before you hand over a credit card.

I've grouped tools by what they're mainly used for, such as text, image, and video, and noted where each company is headquartered, since geography turns out to matter more than you'd think: European and Chinese platforms are undercutting their American counterparts on price almost across the board.

Quick AI pricing comparison

Text and chat assistants

Image generation tools

Video generation tools

Voice AI

Text and chat assistants

ChatGPT — OpenAI (United States)

ChatGPT remains the default AI app for most people, and in 2026, that defaultness comes with the widest plan lineup in the industry. The free and Go tiers now carry ads for users in select markets below each response, a change OpenAI rolled out earlier this year, and Free is genuinely limited to a handful of messages every few hours. 

Plus, at $20 a month, it's still the plan most individual users land on — it adds the company's reasoning model, expanded image generation, limited deep research, and the Codex coding agent.

OpenAI offers two separate Pro price points — $100 and $200 a month — that split usage limits at 5x and 20x, respectively, with the cheaper one positioned against Anthropic's Claude Max. Business pricing is flat per seat, and Enterprise is custom. Given how often this tier structure has shifted this year, treat the exact dollar figures as directional and confirm them at checkout.

Pricing snapshot: Free ($0, ad-supported); Go ($8/month, ad-supported); Plus ($20/month); Pro (at roughly $100–$200/month depending on usage tier); Business ($20/user/month or $25 per user per month when billed monthly); Enterprise (custom).

For API pricing: The GPT-5.5 line, released in April, saw input prices double from $2.50 to $5.00 per million tokens. 

The full flagship lineup as of June 2026: 

• GPT-5.5 runs $5.00 input/$30.00 output per 1M tokens (standard short context), with cached input at $0.50. 
• GPT-5.4 sits at $2.50 input/$15.00 output, while the mini version drops to $0.75 input/$4.50 output. 
• At the top end, GPT-5.5-Pro commands $30.00 input and $180.00 output per 1M tokens. 

Regional processing (data residency) endpoints carry a 10% uplift for models released on or after March 5, 2026. OpenAI models accessed through Amazon Bedrock are billed through AWS and may differ from direct pricing.

Claude — Anthropic (United States)

Claude has stayed the most disciplined of the big three on pricing, holding its Pro tier at a flat $20 a month since launch, with no ad tier and no surprise renewal jump. 

What's changed is everything above Pro. Anthropic now splits its power user option into Max 5x and Max 20x, priced at $100 and $200 a month, which simply multiply how much usage you get per session rather than unlocking new models, a far cleaner structure than rivals' overlapping tiers. 

Claude Pro and Max both bundle access to Claude Code, the company's terminal-based coding agent, at no extra charge, making Claude unusually popular with developers who'd otherwise need a separate coding subscription.

For teams, Anthropic offers Standard seats at roughly $25 a month and Premium seats at around $125 a month when billed monthly, with a discount for annual billing. Enterprise is negotiated directly and provides data residency and security controls that the consumer tiers don't offer. 

Anthropic has been candid that prices and limits can change at its discretion, and the company recently said it's adding substantial new computing capacity specifically to ease usage limits for paying subscribers, a tacit admission that capacity, not just price, has been the real constraint this year.

Pricing snapshot: Free ($0); Pro ($20/month); Max 5x ($100/month); Max 20x ($200/month); Team Standard ($25/seat/month); Team Premium ($125/seat/month); Enterprise (custom).

Anthropic's Claude family offers several current-generation models at distinct price points. 

• Claude Opus 4.7, launched in April, costs $5 per 1M input tokens and $25 per 1M output tokens. 
• Claude Sonnet 4.6 runs $3 input/$15 output per 1M tokens. 
• The budget-friendly Claude Haiku 4.5 is $1 input/$5 output. 
• On the high end, Claude Fable 5 and Claude Mythos 5 cost $10 input/$50 output. 

Gemini — Google (United States)

Google has quietly run the same playbook as OpenAI: a low-cost entry tier, a $20-ish mid-tier most people actually buy, and a top tier that recently split into two price points. Google AI Plus ($4.99 per month) is the budget option for people who mostly want help in Gmail and Docs. 

Google AI Pro, at roughly $19.99 a month, is the closest equivalent to ChatGPT Plus or Claude Pro and bundles Gemini's flagship model with NotebookLM, Jules, and the Flow video-creation tool. Google AI Ultra used to be one $250-ish tier; it's now two — about $99.99 a month for 5x the usage of Pro, and $199.99 a month for 20x, mirroring the exact structure Anthropic and OpenAI have landed on.

The strategic edge Google has that its rivals don't is distribution: a Gemini subscription rides atop Google One storage, YouTube Premium perks, and native integration with Gmail, Docs, and Sheets that a standalone chatbot can't match. 

That's also the catch: Much of Gemini's value is locked to people already in Google's ecosystem, and Workspace business customers buy access through a separate add-on rather than through these consumer plans.

Pricing snapshot: Google AI Plus (entry tier at $4.99); Google AI Pro ($19.99/month); Google AI Ultra 5x ($99.99/month); Google AI Ultra 20x ($199.99/month); Gemini Enterprise (custom, via Workspace).

For developers, Google's Gemini lineup has been aggressive on pricing. Gemini 3.5 Flash, which went GA in May, costs $1.50 per 1M tokens for input and $9.00 for output, with cached input at just $0.15. Gemini API free tier offers 1,500 requests per day. 

Microsoft Copilot (United States)

Copilot is less a single product than a feature bolted onto everything Microsoft sells, which makes it the hardest tool on this list to price cleanly. 

For individuals, Copilot's best AI features now live inside Microsoft 365 Premium, at $19.99 a month or $199.99 a year, layered on top of the existing Personal and Family plans rather than sold as a standalone $20 Copilot Pro subscription, as it once was. 

For businesses, Microsoft 365 Copilot Business is a per-seat add-on to an existing Business plan, list-priced around $21 a seat per month when billed annually (currently discounted at $18/seat/month) or $25.20 per user per month, while larger customers buying Microsoft 365 Copilot at the Enterprise tier negotiate pricing directly tied to their existing E3 or E5 licensing.

The upside of Copilot's approach is that it doesn't require a separate app or login if your organization already runs on Microsoft 365 — Copilot just shows up inside Word, Excel, Outlook, and Teams. The downside is that pricing is genuinely fragmented across at least four different purchasing paths (Personal, Premium, Business, Enterprise), and Microsoft has been adjusting list prices and bundling rules every few months, including a global pricing update planned for July 2026.

Pricing snapshot: Microsoft 365 Premium with Copilot ($19.99/month or $199.99/year, individuals); Copilot Business add-on ($21/seat/month, list price); Copilot Enterprise (custom, tied to M365 E3/E5).

Perplexity (United States)

Perplexity has built its identity around cited, sourced answers rather than raw chat, and its pricing ladder reflects a company trying to serve everyone, from students to enterprise research teams, off the same core product. 

Free gives a handful of advanced searches a day. Pro, at $17 a month or $200/year, removes that cap and adds access to multiple underlying models — OpenAI's, Anthropic's, and its own Sonar models — switchable inside one interface, which is arguably Perplexity's sharpest pitch: one subscription instead of three. 

Max, introduced as the company's flagship power tier, costs a flat $200 per month or $2,000 annually and adds an agentic system called Perplexity Computer, the company's Comet browser's most advanced assistant mode, and far higher limits on report- and dashboard-generation tools.

On the business side, Enterprise Pro runs about $40 a seat per month and Enterprise Max around $325 a seat per month, a steep jump justified mainly by security features like SCIM and audit logs that only unlock at scale. 

Education users get 50% off the Pro tier. The honest read on Perplexity's pricing is that Pro is genuinely competitive at $20, but Max's $200 price tag asks regular consumers to pay the same as agencies, with little in between.

Pricing snapshot: Free ($0); Pro ($17/month or $200/year); Max ($200/month or $2,000/year); Education Pro ($10/month); Enterprise Pro ($40/seat/month); Enterprise Max ($325/seat/month).

Le Chat/Vibe — Mistral AI (France)

Mistral is Europe's most credible answer to the American labs, and its assistant — recently rebranded from Le Chat to Vibe — undercuts nearly every Western rival on price while folding in a genuinely useful coding agent most competitors charge separately for. 

The free tier already includes web search, image generation, and basic coding help through Vibe for code, which is more generous than most free AI chat tiers. 

Pro costs $14.99 a month, a deliberate few dollars below the $20 that's become the industry's unofficial standard, and verified students pay just $5.99.

Mistral's pitch leans hard on European data sovereignty and self-hosting flexibility; its open-weight models can be run on a company's own infrastructure under a commercial license, an option none of the big three American labs offer in the same way. Team plan requires a minimum of two users and costs $24.99 per user per month.

Enterprise pricing isn't published and requires contacting sales, which is standard for the category, but means European businesses evaluating Mistral against Microsoft or Google still need a sales call to get a real number.

Pricing snapshot: Free ($0); Pro ($14.99/month, $5.99/month for verified students); Team ($24.99/user/month); Enterprise (custom, contact sales).

The API and Le Chat/Vibe consumer interface run on completely separate billing systems. Mistral Large 3, the current flagship, costs $0.50 per 1M tokens for input and $1.50 for output. Mistral Small 4 runs $0.15 input/$0.60 output, making it a strong production default. At the budget end, Mistral NeMo is just $0.15 input/$0.15 output. Reasoning models (Magistral Medium) cost $2.00 input/$5.00 output.

DeepSeek (China)

DeepSeek's entire pitch is that frontier-adjacent AI doesn't need frontier pricing, and on the API side, the numbers back that up dramatically. The consumer chat app and mobile apps are completely free with no subscription tier at all, a structure no major Western lab has matched. 

The real story is the developer API: DeepSeek's V4 Flash model charges $0.14 per million input tokens and $0.28 per million output tokens, which independent trackers peg at roughly 95% to 99% cheaper than GPT-5-class or Claude-class API pricing for comparable workloads. 

The flagship V4 Pro reasoning model costs more — list price around $1.74 input and $3.48 output per million tokens — though DeepSeek has been running a steep promotional discount ($0.435 input/$0.87 output) on that tier through most of 2026.

The catch isn't the price; it's everything around it. DeepSeek's infrastructure runs primarily out of China, which means variable latency for users elsewhere and content restrictions on politically sensitive topics baked into both the chat app and, to a lesser extent, the API. 

For cost-sensitive, high-volume workloads where censorship and data residency aren't dealbreakers, DeepSeek is simply the cheapest credible option on the market; for regulated industries or anyone needing guaranteed uptime, most engineering teams still route through a Western reseller for redundancy.

Pricing snapshot: Chat app (free, no tiers); API — V4 Flash ($0.14 input/$0.28 output per 1M tokens); V4 Pro ($1.74 input/$3.48 output per 1M tokens, list price; heavily discounted through much of 2026).

Image generation tools

Midjourney (United States)

Midjourney is still the aesthetic benchmark the rest of the image-generation field is measured against, and it's also the one major platform that has never offered a free tier; you pay for GPU time from the first image. 

The structure is refreshingly simple compared to the credit chaos elsewhere on this list: four flat tiers — Basic, Standard, Pro, and Mega — priced at $10, $30, $60, and $120 a month, each buying a set number of hours of "Fast" generation time, with unlimited slower "Relax" generation included from Standard upward. Annual billing knocks 20% off any tier.

The tradeoff for that simplicity is accessibility. Midjourney still runs primarily through Discord, which remains a genuine barrier for people who don't already use the chat app, even though a web interface now exists. Stealth mode — keeping your generations private from the public feed — is locked behind the Pro and Mega tiers, and any company earning more than $1 million a year in revenue is contractually required to be on one of those two top plans to retain ownership of what it generates.

Pricing snapshot: Basic ($10/month); Standard ($30/month); Pro ($60/month); Mega ($120/month) — all with a 20% discount on annual billing.

Adobe Firefly (United States)

Firefly is Adobe's bet that "commercially safe," trained on licensed Adobe Stock content rather than scraped web images, is worth paying a premium for, and its pricing reflects a company trying to upsell its way through an existing Creative Cloud customer base. 

The cheapest standalone option, Firefly Standard, runs $9.99 a month for 2,000 monthly credits and unlimited use of Adobe's standard image and vector tools; credits are only spent on premium features like video generation and third-party model access (Google's Nano Banana, OpenAI's image model, and others Adobe has folded into the same interface). 

Firefly Pro roughly doubles the credit pool for around $19.99, while Firefly Pro Plus for teams costs $49.99 per license per month, billed annually, and Firefly Premium tops out at $199.99 a month for high-volume studio use.

The wrinkle is that Adobe also gives away a separate pool of generative credits to anyone already paying for Creative Cloud Pro, which now includes 4,000 monthly credits as part of its base subscription, meaning plenty of Photoshop and Illustrator subscribers may not need a standalone Firefly plan at all. Adobe has also been running unlimited-generation promotions on its paid tiers periodically throughout 2026, which makes the credit math a moving target.

Pricing snapshot: Free (limited credits, watermarked); Standard ($9.99/month, 2,000 credits); Pro ($19.99/month, 4,000 credits); Pro Plus ($49.99/month, 10,000 credits); Premium ($199.99/month, 50,000 credits).

Leonardo AI (Australia)

Leonardo built its following on giving creators more granular control than Midjourney, custom model training, a real-time canvas, and a generous free tier that, unlike Midjourney's, costs nothing to try. 

It's also the rare major image tool headquartered outside the US and Europe. The free plan provides 150 daily credits, enough for a couple of dozen casual images, though everything generated on it is public and Leonardo retains broad usage rights. 

Paid tiers — historically called Apprentice, Artisan, and Maestro and recently rebranded by Leonardo to Essential, Premium, and Ultimate — run $12, $30, and $60 a month, each roughly tripling the monthly token pool, with annual billing cutting the effective price by about 20%.

The catch that catches people off guard is that "unlimited" only applies to Leonardo's own models, like Phoenix and Lucid Origin; every time you reach for a premium third-party model, such as FLUX, Veo 3, Kling, Seedream, or GPT-Image-1.5, it draws down your token balance like everything else. Leonardo recently shifted its developer API entirely to pay-as-you-go billing, a sign the company expects usage-based pricing to eventually replace flat subscriptions even on the consumer side.

Pricing snapshot: Free (150 daily tokens); Apprentice/Essential ($12/month); Artisan/Premium ($30/month); Maestro/Ultimate ($60/month); Teams (from ~$24/seat/month).

Stable Diffusion/Stable Assistant — Stability AI (United Kingdom)

Stability AI occupies an unusual spot in this market: it's best known for giving away open-weight models that anyone can download and run for free, while also selling its own polished consumer products on top of them. 

Stable Assistant, the company’s chatbot and image generator built around Stable Diffusion 3.5, starts at $9 a month with a three-day free trial, billed via a credit system where a successful image generation costs roughly 6.5 credits. Stable Artisan, a Discord-native version of the same technology, follows the same starting price.

For developers, the more relevant product is the API, billed at a flat $0.01 per credit with no subscription required, genuinely simple compared to the subscription-plus-credits hybrids most rivals run. The bigger strategic point is that Stability's open licensing means a company with the engineering resources to self-host can skip the subscription question entirely, something that's not realistically true of Midjourney, Adobe, or Leonardo. 

That openness is also the source of most of Stability's well-documented commercial turbulence over the past few years, as the company has repeatedly had to balance giving models away against building a sustainable paid business around them.

Pricing snapshot: Standard plan (from $9/month, 900 credits monthly); Pro ($19/month, 1,900 credits monthly); Plus ($49/month, 5,500 credits monthly); Premium ($99/month, 12,000 credits monthly). All plans include a 3-day free trial.

Developer API (pay-as-you-go, $0.01 per credit); self-hosted open-weight models (free for research, licensed for commercial use).

Video generation tools

Runway (United States)

Runway is the platform that took AI video from a novelty to something agencies actually bill clients for, and its 2026 pricing overhaul replaced a confusing "Unlimited" tier with a cleaner Max plan. 

The structure now runs Free (a one-time, non-renewing 125 credits), Standard ($12 a month annually, $15 month to month), Pro ($28 annually, $35 monthly), and Max ($76 annually, $95 monthly), each tier roughly multiplying the monthly credit pool while keeping the same per-second cost across models. 

Every paid tier includes Runway's own Gen-4.5 model, along with licensed access to Google's Veo, Kling, and other third-party video models, all within the same interface.

The credit math is where Runway gets genuinely confusing: a second of footage from the flagship Gen-4.5 model costs five times as many credits as the same second from the cheaper Gen-4 Turbo model, so the headline "625 credits a month" on Standard translates to anywhere from 25 to 125 seconds of usable video depending entirely on which model you pick. 

Pricing snapshot: Free (125 one-time credits); Standard ($12/month annual, $15 monthly); Pro ($28/month annual, $35 monthly); Max ($76/month annual, $95 monthly); Enterprise (custom).

Synthesia (United Kingdom)

Synthesia took a different bet than every other video tool on this list: instead of generating cinematic clips from text prompts, it builds talking AI avatars for corporate training, onboarding, and explainer videos, and it's the only platform here that prices by minutes of finished video rather than generation credits. 

Starter costs $14 per month when billed annually, or $19 per month. The Creator plan costs $39 per month when billed annually, or $49 per month. A free plan is available, but it caps at just 10 minutes per month.

Synthesia's customer base skews heavily toward corporate learning-and-development and internal communications teams rather than individual creators, which shows up in the pricing: Enterprise is the only tier with unlimited minutes, custom avatars at scale, and the security certifications that procurement teams at large companies actually ask for. 

For a business already paying agencies or studios to produce training videos, Synthesia's per-minute model, even at the Creator tier, is cheaper than traditional production, which is precisely the pitch.

Pricing snapshot: Free (10 minutes/month); Starter ($14/month annual, $19 monthly, 120 minutes/year); Creator ($39/month annual, $49 monthly, 360 minutes/year); Enterprise (custom, unlimited).

Luma AI (United States)

Luma built its reputation on Dream Machine, a video model praised for unusually realistic physics. Luma AI’s Dream Machine focuses on cinematic realism, tracking physical camera movements, and maintaining strict character consistency across video cuts.

Luma is priced at Plus ($30 per month), Pro ($90 per month), and Ultra ($300 per month). The company raised a large funding round ($900M) in late 2025 at a multibillion-dollar valuation, which makes a near-term price cut unlikely.

Pricing snapshot: Plus ($30/month); Pro ($90/month); Ultra ($300/month). 

Kling AI — Kuaishou (China)

Kling, built by Chinese tech company Kuaishou, has climbed near the top of independent video-model leaderboards in 2026 while undercutting every Western competitor on entry price. 

Its free tier hands out 66 credits daily, enough for one or two short clips, and the cheapest paid tier, Standard, runs as low as $6.99 a month for new subscribers (renewing at roughly $8.80), making it arguably the lowest-priced commercial-use AI video plan on the market. Pro runs about $25.99 a month, Premier about $64.99, and the top Ultra tier, which jumped from roughly $128 to $180 a month in a price increase earlier this year, provides the largest credit pool and earliest access to Kling's newest model.

Kling's standout technical feature is native audio generation with lip-sync built directly into the model, something most rivals still treat as a separate, more expensive add-on. The tradeoffs mirror DeepSeek's: data processing happens primarily in China, and customer support is thin by Western SaaS standards.

Pricing snapshot: Free (66 daily credits); Standard (from $6.99/month intro, ~$8.80 renewal); Pro (~$25.99/month); Premier (~$64.99/month); Ultra (~$180/month).

Pika (United States)

Pika carved out a niche with playful, effects-driven video tools that let creators swap objects into real footage or apply cartoonish physics, leaning more toward social-media virality than cinematic realism. 

Its four tiers run Free, Standard ($8 per month annually, $10 monthly), Pro ($28 per month annually, $35 monthly), and Fancy ($76 per month annually, $95 monthly), and the company is unusually upfront that Standard — despite being a paid plan — still doesn't include commercial usage rights or watermark-free downloads; both only unlock starting at Pro.

That detail trips up many new subscribers who assume any paid tier removes restrictions, and it makes Pika's real "starting price" for commercial use $28 per month rather than the $8 headline figure. On the upside, Pika is one of the only platforms here where separately purchased top-up credits never expire, unlike monthly plan credits, giving heavier users a way to bank capacity for busy months, a small but genuinely useful flexibility most competitors don't offer.

Pricing snapshot: Free ($0, watermarked, no commercial use); Standard ($8/month annual, $10 monthly, no commercial use); Pro ($28/month annual, $35 monthly, commercial use included); Fancy ($76/month annual, $95 monthly).

Voice AI

ElevenLabs

ElevenLabs is an industry standard for realistic synthesis, voice cloning, and long-form speech engineering. It operates across a clear, volume-based character credit matrix. ElevenLabs increasingly dominates AI voice generation because its output quality is consistently strong across narration, voiceovers, and multilingual speech.

Voice cloning remains one of its biggest strengths, though the credit system can make monthly spending harder to predict at scale. The Starter plan costs $6 per month for 30,000 credits and basic voice cloning. 

The Creator plan scales to $22 per month for 121,000 credits and unlocks Professional Voice Cloning. Higher-volume requirements move through the Pro tier at $99 per month for 600,000 credits, up to the Scale tier at $299 per month for 1.8 million credits.

Pricing snapshot: Free ($0, 10K credits/month); Starter ($6/month, $5/month/annually); Creator ($22/month; $11 for the first month with 50% off; $18.33/month annually); Pro ($99/month, $82.5/month/annually); Scale ($299/month, $248.17/month/annually); Business ($990/month, $825/month/annually); Enterprise (custom pricing).

What the numbers actually tell us

Step back from the individual tools, and a few patterns jump out. The most obvious is convergence: ChatGPT Plus, Claude Pro, Gemini Pro, and Perplexity Pro have all landed within a dollar of $20 a month, which has effectively become the industry's default price for serious individual use, the same way $9.99 became the default for a single streaming service a decade ago. 

That's not really competition driving prices down; it's four well-funded companies independently arriving at the same number and refusing to move it, even as the underlying models get more capable each year.

The second pattern is the quiet split of top-tier plans into two price points instead of one. Anthropic did it first with Max 5x and 20x; Google followed with two Ultra tiers; and OpenAI followed the same structure for its Pro plan. 

That's a meaningful shift from simply raising the price of one plan; it suggests these companies have concluded that usage limits, not model access, are now the thing power users will actually pay extra for, because the flagship models are now bundled at every paid tier rather than gated behind the most expensive one.

The third, and arguably most consequential, pattern is geographic. Every Chinese platform on this list — DeepSeek and Kling — undercuts its closest Western equivalent by an order of magnitude or more on a like-for-like basis, whether that's API token pricing or a monthly video subscription. 

Europe's major entrant here, Mistral, doesn't undercut by nearly as much, but still prices a few dollars below the American $20 standard, emphasizing data sovereignty as the differentiator rather than raw cost. 

Whether the Chinese pricing reflects genuinely lower compute costs, government-subsidized infrastructure, or a deliberate land-grab strategy is a real open question, but the price gap itself isn't in dispute.

Also read: This AI cheat sheet explains major AI tools, free options, chatbot categories, humanoid robots, and the terms buyers keep seeing in 2026.