Claude vs ChatGPT Pricing Compared: Free and Paid Plans to API Costs

"I want to use an AI assistant, but should I pay for Claude or ChatGPT?" "How far can I get on a free plan?" Plenty of people are asking the same thing. As of October 2026, both companies offer individual plans at $20, $100 and $200 (ChatGPT also has a $500 tier), and even their team seats cost the same. So the differences show up elsewhere: in the free and $8 entry points, in how much of the top model the upper plans give you, and in the conditions attached to API rates.

In this article, we compare Claude (Anthropic) and ChatGPT (OpenAI) pricing across five angles: free, paid subscriptions, premium plans, teams and the API. Unless noted otherwise, prices are the official US-dollar figures. We checked the plans on September 22, 2026 (rechecking ChatGPT Pro and the chat models on October 12) and the API on October 12, using each company's pricing pages and help center (for local-currency display, see the FAQ). Whether Anthropic's IPO affects Claude's pricing and plans is covered in Anthropic IPO: Date, S-1 Details, and Whether You Can Buy.

1. Pricing Overview

Let's start with the big picture. Both services follow the same ladder (free, individual paid, power user, team, enterprise), and as of September 2026 their price points are almost identical too.

Sources: Claude pricing page / ChatGPT pricing page, Codex "Pricing", OpenAI Help "About ChatGPT Pro tiers" (checked September 22, 2026; Pro tiers rechecked November 2). The Pro multipliers are our calculation: we applied the October 29 allowance increase (help center article above) to the multipliers in a post by Tibo, who leads Codex and ChatGPT at OpenAI (September 29, 2026), and OpenAI's DevDay announcement. For team seats, the lower price is for annual billing and the higher one for monthly billing.

The differences are at the two ends of the ladder. At the bottom, only ChatGPT has the $8/month Go plan; Claude has no paid plan below $20. At the top, both offer a $100 plan, but with different multipliers (Claude's Max gives 5x Pro, ChatGPT's Pro 100 gives 10x Plus). Claude stops at Ultra ($200; renamed from Max 20x on October 31, 2026), while ChatGPT Pro has had three tiers, $100, $200 and $500, since September 29, 2026 (new sign-ups for Pro $200, paused since September 10, reopened the same day). On October 29, the Work and Codex allowances on Pro 100 and Pro 500 doubled, and everyone on Pro 200 moved to the higher allowance, so with Plus as 1 the tiers give 10x, 20x and 50x (our calculation, applying the October 29 change to the multipliers OpenAI staff and OpenAI announced on September 29; the Plus allowance is unchanged).

2. Free Plan Comparison

If you want to try things out for free first, the free plans matter a lot. This is also where the two companies' different approaches show most clearly.

Sources: Claude pricing page / ChatGPT pricing page, ChatGPT release notes (October 7, 2026 entry), OpenAI Help "GPT-6 and other models in ChatGPT" (checked October 26), OpenAI Help "Ads in ChatGPT" (checked September 22, 2026), OpenAI "ChatGPT Ads expands to Southeast Asia and Taiwan" (September 23, 2026; checked September 29), OpenAI "ChatGPT Ads expands to 26 new countries" (October 21, 2026; checked October 26)

Key point: ChatGPT's free plan lets you chat in text as much as you like on the lightweight GPT-6 Luna, and you can also try image generation and Codex. OpenAI's release notes and pricing page say Free and Go are moving to GPT-6 Luna in a rollout starting October 8, 2026, but its model help page still lists GPT-5.6 Luna for Free and Go as of October 26, so check the model name on your own screen. On the other hand, OpenAI began testing ads in the US on February 9, 2026, and on September 23 expanded them to Southeast Asia and Taiwan (in Asia-Pacific, Japan, South Korea, India, Australia and New Zealand came earlier). On October 21, 26 more countries, including several in Latin America, were added, bringing ads to 89 countries. Ads appear only on Free and Go; Plus, Pro and Enterprise stay ad-free (as of the October 21, 2026 announcement). Anthropic stated on February 4, 2026 that Claude will remain ad-free.

Claude's free plan hits its usage limit sooner, but it lets you try Sonnet, Anthropic's mid-range tier (ChatGPT's free plan runs on Luna, which OpenAI positions as its low-cost, high-volume tier). However, Claude Code is not included in the free plan, so Codex is the only coding agent you can try without paying.

3. Paid Subscription Comparison

Here are the main plans for individual users. Both cost $20 a month.

Sources: Claude pricing page, Claude Code "Model configuration" (rechecked September 25) / ChatGPT pricing page, OpenAI Help "GPT-6 and other models in ChatGPT", "About ChatGPT Pro tiers" (checked September 22, 2026; chat models and context rechecked October 12)

On the ChatGPT chat screen, you pick a model and then a thinking level (Instant, Medium, High and so on). Since October 7, 2026, chat defaults to GPT-6, and according to OpenAI's help center all of these levels run on GPT-6 Sol for Plus users (before that it was GPT-5.6 Sol, which you can still select explicitly). GPT-6 Astra, released September 3, 2026, does not appear in Plus chat. On Plus, Astra is available only inside ChatGPT Work and Codex, and usage is capped.

Claude Pro lets you choose Opus or Sonnet both in chat and in Claude Code. Claude Code's default model changes each time a new model ships: since the Opus 5.5 update on September 22, 2026 (v2.1.280), it is Opus 5.5 on both Pro and Max (before that, Pro defaulted to Sonnet 5 and Max to Opus 5, per the official Claude Code documentation). The top model, Fable 5.1, isn't covered by the Pro subscription's included usage; you can use it only by paying for extra usage credits.

The same $20 a month gets you different strengths. Claude shines when you want to feed it long documents or codebases in one go (up to 1M tokens) and for coding (Claude Code). ChatGPT stands out for image generation and for letting you try OpenAI's top model, GPT-6 Astra, in limited amounts inside Work and Codex. For how to use each of Claude's features, see "Claude's Chat, Cowork and Code: how they differ and when to use each."

4. Premium Plan Comparison (Max vs Pro)

For people who use AI for hours every day, Claude offers two tiers ($100 and $200) and ChatGPT three ($100, $200 and $500). The differences are the multiplier at $200, how much of the top model you get, and the $500 tier that only ChatGPT has.

Sources: Claude pricing page, Claude Code "Model configuration" (rechecked September 25) / OpenAI Help "About ChatGPT Pro tiers", "GPT-6 and other models in ChatGPT", ChatGPT pricing page (rechecked October 12, 2026), OpenAI staff post ($100 and $200 multipliers), DevDay announcement ($500 multiplier). The Pro multipliers are our calculation, applying the October 29 allowance increase (About ChatGPT Pro tiers, checked November 2) to these two

If $20 isn't enough but $200 feels like too much, both companies have a $100 plan. The multipliers are 5x Pro for Claude's Max and 10x Plus for ChatGPT's Pro 100 (since October 29, 2026). Anthropic also said in its Opus 5.5 announcement (September 22, 2026) that it had raised the 5-hour usage limits for Pro, Max, Team and seat-based Enterprise, but it did not say by how much. Prices and multipliers are unchanged.

On ChatGPT Pro, GPT-6 Pro (which runs on GPT-6 Astra) comes as a message allowance, and OpenAI's help center states plainly that a Pro subscription does not make GPT-6 Pro unlimited. At $100, GPT-6 Pro and GPT-5.6 Sol Pro share one weekly allowance; at $200, once you use up the GPT-6 Pro weekly allowance, ChatGPT switches to GPT-5.6 Thinking (Medium). The "50 a week" and "200 a week" counts we saw in the help center on September 22 no longer appear there as of October 12. On Claude's Max and Ultra, you can spend up to half of your weekly usage limit on Fable 5.1. Both $200 plans are open to new sign-ups; Claude Ultra gives 10x Pro per week (no 5-hour limit if verified), and ChatGPT Pro $200 gives 20x Plus (since October 29, 2026). Since the base plans' allowances differ between the two companies, the multipliers alone don't tell you which gives more usage. ChatGPT also has Pro 500 above it ($500, 50x Plus), the only Pro tier where you can use Ultrafast (the faster speed setting) for GPT-6 Astra and GPT-6.1 Sol.

5. Team and Business Plan Comparison

Here's how the plans compare for organizations. As of September 2026, the two companies' team plans are priced in almost exactly the same way.

Sources: Claude pricing page (Team & Enterprise) / OpenAI Help "ChatGPT Business - Overview", "GPT-6 and other models in ChatGPT" (checked September 22, 2026)

What is a "premium seat"? It's a higher tier of seat, separate from the regular one, with more usage. Both companies let you mix standard and premium seats in a single workspace, so you can give premium seats only to developers and other heavy users.

With identical prices, the deciding factors are which one your organization already uses and how you want to use the top model. On both plans, conversation data is not used for AI training by default, which makes either one easier for companies to adopt with confidence.

6. API Pricing Comparison

For developers, API pricing matters most. Both companies offer a recommended starting point ("if in doubt, use this"), a mid-range tier that balances speed and price, and a low-cost tier for high-volume work.

Sources: Anthropic "Pricing", "Models overview" / OpenAI "Pricing", "Models" and the individual model pages (checked October 12, 2026). Positioning follows each company's official model list. The three OpenAI models in the table accept at most 922K input tokens per request. Haiku 5.5's rates apply to prompts of up to 100K tokens; above that, they rise to $0.50 input, $0.05 cache reads and $2.50 output.

Both companies open their official model lists with an "if in doubt, start here" recommendation, and what it points to changes with every new release. Where they differ is how the lineup is built. OpenAI makes the starting point and the top model the same thing (as of September 2026, GPT-6 Astra). Anthropic keeps them separate: at the same point, its guidance is to start with Opus 5.5 for most use cases and turn to Fable 5.1 for demanding reasoning, long-running agentic work, or when Opus 5.5 isn't enough. In other words, if you compare each company's starting point, you're comparing Anthropic's cheaper tier against OpenAI's top one.

The previous generation, now out of the table, isn't disappearing right away on either side. Anthropic's Opus 5 ($5/$25); Sonnet 5 ($2/$10, the same input and output rates as Sonnet 5.5, with cache reads at $0.20, twice Sonnet 5.5's), which Sonnet 5.5 replaced on September 28; and Haiku 4.5 ($1/$5), which Haiku 5.5 replaced on October 7, are now listed as Legacy (still available) in the official list. OpenAI's GPT-5.6 Terra ($2/$12) and GPT-5.6 Luna ($0.20/$1.20; rates checked September 22) remain in the API model list. GPT-5.6 Sol ran chat on ChatGPT's paid plans until GPT-6 Sol replaced it on October 7 (it can still be selected in chat), and on the API its promotional price of $4/$20 runs until at least November 21, 2026 (the regular price is $5/$30).

OpenAI's mid-range model in the table is GPT-6.1 Sol, released September 29 (US time). The model OpenAI's API model list recommends for balancing intelligence and cost changed from GPT-6 Sol to this one (checked September 30, 2026). Compared with GPT-6 Sol, the $2 input and $10 output rates and the way the long-input surcharge applies are the same; only cache reads changed, to $0.10, half the price (5% of the input rate). Watch where you can use it: in ChatGPT it is available in Work and Codex (at launch for Plus, Pro, Business, Enterprise and Edu; Free and Go are not included), and not in Chat. GPT-6 Sol ($2/$0.20/$10) also remains in the API.

Conditions the table doesn't show: long inputs, caching and token counts

• Long-input surcharge: For the three OpenAI models in the table (and GPT-5.6 Sol), once input exceeds 272K tokens, the entire request is billed at 2x for input (and cache) and 1.5x for output. Claude models from Claude 4.6 onward stay at standard rates all the way to 1M tokens, except Haiku 5.5: once a prompt exceeds 100K tokens, its input, cache and output rates all rise 5x ($0.50 input, $2.50 output)
• Caching: Cache reads cost 10% of the input rate on both sides by default (on Claude, 5% for Opus 5.5 and Sonnet 5.5 and 2.5% for Fable 5.1; on OpenAI, 5% for GPT-6.1 Sol). Cache writes cost 1.25x the input rate on both (for Claude, that's the 5-minute cache; the 1-hour cache is 2x)
• Batch: 50% off both input and output on both sides. OpenAI also offers Flex processing, which tolerates longer waits, at the same rate
• Fast mode: Claude's fast mode (Opus only; Opus 5.5 leaves the research preview on October 15, 2026 and no longer requires an application, while Opus 5 and 4.8 remain in research preview) costs 2x standard ($8/$40 for Opus 5.5); OpenAI's Fast mode is also 2x ($20/$100 for Astra)
• Regional processing: Claude charges 1.1x for US-only processing (models from Claude 4.6 onward); OpenAI charges 10% more for regional processing (eligible models released on or after March 5, 2026)
• Token counts themselves differ: According to Anthropic, the new tokenizer in models from Claude 4.7 onward (Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1 and others) counts about 30% more tokens for the same text (1.0–1.35x depending on content, per the Sonnet 5 announcement). The Legacy Haiku 4.5 still uses the previous tokenizer, so moving to Haiku 5.5 raises the token count for the same text. OpenAI counts differently as well, so to compare rates, count the tokens on your own text

What it comes to per request

A table of rates alone doesn't show where the gaps open up. We calculated the cost of sending the same request once, assuming identical token counts (standard rates, no caching or batch).

Example calculation: Sonnet 5.5, long document = 400K × $2 ÷ 1M + 10K × $10 ÷ 1M = $0.80 + $0.10. GPT-6.1 Sol exceeds 272K tokens, so it's calculated at $4 input and $15 output (400K × $4 ÷ 1M + 10K × $15 ÷ 1M = $1.60 + $0.15). Haiku 5.5 exceeds 100K tokens, so it's calculated at $0.50 input and $2.50 output (400K × $0.50 ÷ 1M + 10K × $2.50 ÷ 1M = $0.20 + $0.025). Rates come from the same sources as the table above.

The table above assumes no caching. Even with caching, the two mid-range models come out even. For a request that resends the same context, so that 90K of its 100K input tokens are cache reads (plus 2K output), Sonnet 5.5 costs 10K × $2 + 90K × $0.10 + 2K × $10 (each ÷ 1M) = $0.020 + $0.009 + $0.020 = $0.049, and GPT-6.1 Sol comes to the same $0.049 (with GPT-6 Sol, or with Sonnet 5.5's $0.20 cache reads through October 6, it would be $0.058). The initial cache write is not included.

API pricing bottom line: which one is cheaper flips depending on the tier and the input length.

• Top tier vs. top tier (Fable 5.1 and Astra): the rates are identical. But Astra carries a surcharge on long inputs, and its cache reads cost 4x as much as Fable 5.1's ($1 vs. $0.25), so Fable 5.1 comes out cheaper for long documents or workflows that reuse the same context
• Starting point vs. starting point: Opus 5.5 ($4/$20) costs two-fifths of Astra ($10/$50). But as noted above, OpenAI's starting point is also its top model, so this isn't a like-for-like tier comparison. Against OpenAI's mid-range GPT-6.1 Sol ($2/$10), Opus 5.5 is twice as expensive
• Balanced: as of October 12, 2026, Sonnet 5.5 and GPT-6.1 Sol have the same input, cache-read and output rates ($2/$0.10/$10). For inputs up to 272K tokens there's no difference, with or without caching; for inputs over 272K tokens, Sol's surcharge makes Sonnet 5.5 roughly half the price (our September 22 version said "Sonnet 5 is cheaper on output," which stopped being true when the comparison switched from GPT-5.6 Terra, at $12 output, to GPT-6 Sol; and our September 30 version said "the more of your input hits the cache, the cheaper GPT-6.1 Sol gets," which stopped being true when Sonnet 5.5's cache reads dropped from $0.20 to $0.10 on October 7). Note that GPT-6.1 Sol also has Ultrafast pricing on the API, which runs the same model faster: $12 input and $60 output for inputs up to 272K tokens (6x standard) (OpenAI's API pricing page, checked October 18, 2026).
• Low cost: Claude Haiku 5.5, released October 7, and GPT-6 Luna have the same input, cache-read and output rates ($0.10/$0.01/$0.50) for prompts of up to 100K tokens. Above 100K tokens, only Haiku 5.5 rises 5x ($0.50/$2.50), while Luna's surcharge (2x input, 1.5x output) starts only above 272K tokens, so Luna is cheaper on long inputs ($0.088 vs. $0.225 for the 400K-token example above). Against Haiku 4.5 ($1/$5), the model our versions through October 6 compared, Luna was one-tenth the price on both input and output

Some things can't be compared on price alone, because models differ in how many tokens (how much thinking and how many steps) they need to finish the same job. For a detailed comparison of the two for coding, see "Claude Code vs Codex: a full comparison." For choosing among Claude's tiers, see "Claude pricing by model (Opus, Sonnet, Haiku)."

7. Cost-Saving Techniques

Here are the techniques that cut API costs the most.

Saving on Claude

Saving on ChatGPT (OpenAI)

On both sides, combining the Batch API with prompt caching is the most effective approach. If you process large volumes, make full use of both. Pricing sources are the same as for the table in section 6. For how the two providers' caching differs and how to check that it is actually working, see "OpenAI vs Claude Prompt Caching".

8. Recommended Plans by Use Case

Finally, here are our recommendations by how you use AI.

9. FAQ

* The plans in this article were checked on each company's official pages on September 22, 2026 (ChatGPT Pro and the chat models rechecked on October 12), and the API prices on October 12, 2026. Pricing and plan details may change without notice. For the latest information, see the official sites (claude.com, chatgpt.com, Claude API, OpenAI API).