ChatGPT vs Claude: Which Costs Less in 2026?
A precise cost comparison between ChatGPT and Claude — $20 subscriptions on both sides, and per-token API rates — to find the cheaper option for your actual usage.
The article text carries the review date. The rate table below is rebuilt from the live catalogue on every deploy.
Fastest win
ChatGPT and Claude cost the same for individuals — $20/month for ChatGPT Plus or Claude Pro (ChatGPT Go is $8; ChatGPT Business is $25/seat; Claude Max starts at $100). On the API, Claude Sonnet 5 lists $2 input / $10 output per 1M tokens and the current GPT flagship tier sits in the same band, so the saving is in routing routine traffic to a budget row, not in picking a vendor.
Subscription Pricing: Dead Even
At the subscription level, ChatGPT and Claude still cost the same:
- ChatGPT Plus: $20/month
- Claude Pro: $20/month
- ChatGPT Business: $25/user/month
- Claude Team: seat pricing, check the vendor page
Seat prices are vendor-set, so confirm them on the vendor page before you buy — they are not part of the hourly catalogue sync. The subscription decision should turn on features and workflow fit, not price, because the price is identical.
API Pricing: The Input Gap Has Closed
This is where the 2024 advice stopped being true. Flagship API rates are now level:
OpenAI GPT Latest:
- Input: $2.00 per 1M tokens
- Output: $10.00 per 1M tokens
Anthropic Claude Sonnet Latest:
- Input: $2.00 per 1M tokens
- Output: $10.00 per 1M tokens
Input rates are level. Output rates are level too. If you are still routing input-heavy work to Claude because it used to be 40% cheaper on input, that rule has expired — the saving it was built on no longer exists.
Budget tier: OpenAI GPT Mini Latest lists $0.75 input / $4.50 output per 1M, against Anthropic Claude Haiku Latest at $1.00 / $5.00. GPT mini is the cheaper budget row today, but the margin is a fraction of what it once was, so quality on your own prompts decides it rather than the price sheet.
Neither model wins on price — pick on accepted output
The old input-price gap has closed. Claude Sonnet lists $2.00 input / $10.00 output per 1M and GPT-6.1 Sol, OpenAI's high-end row below Astra, lists $2.00 / $10.00. When the rates match, the cheaper model is whichever one needs fewer retries and less human editing for your work.
September 2026 API Rows: Opus 5.5, GPT-6.1 Sol, Luna, Astra and Fable
OpenAI and Anthropic both added API rows in September 2026. Prices below are per 1M tokens as billed on OpenRouter, checked on 1 October 2026; the batch column is OpenRouter's Batch API rate.
| Model | Input / output per 1M | Batch per 1M | What it is for |
|---|---|---|---|
| GPT-6 Luna (22 Sep) | $0.10 / $0.50 | $0.05 / $0.25 | OpenAI's fast tier for high-volume chat and classification |
| GPT-6.1 Sol | $2 / $10 | $1 / $5 | OpenAI's cost-efficient high-end row, below Astra; $4 / $15 once a prompt reaches 272K tokens; cached input $0.10 |
| Claude Opus 5.5 (22 Sep) | $4 / $20 | $2 / $10 | Anthropic's new flagship for reasoning, coding and long-horizon agents; 1M context |
| GPT-6 Astra (4 Sep) | $10 / $50 | $5 / $25 | OpenAI's flagship for long-horizon agentic and computer-use work; 1M context |
| Claude Fable 5.1 (1 Sep) | $10 / $50 | $5 / $25 | Anthropic's row for agentic coding and long-running workflows; 1M context |
GPT-6.1 Sol is the current Sol row; it replaced GPT-6 Sol, which launched on 22 September at the same $2 / $10. Claude Sonnet 5 is still $2 / $10, so the mid-tier comparison above holds: Sol and Sonnet are level on rate.
The top tier is not level. Claude Opus 5.5 costs twice GPT-6.1 Sol on both input and output. GPT-6 Astra and Claude Fable 5.1 are level with each other at $10 / $50, two and a half times Opus 5.5. OpenRouter's Intelligence Index ranked Opus 5.5, Fable 5.1 and Astra as its top three after September, so of the three highest-ranked rows Opus 5.5 is by far the cheapest.
Worked Example: 1M Input + 200K Output, Interactive vs Batch
Assumptions: one workload of 1M input tokens and 200K output tokens, no prompt caching, prompts under 272K tokens, OpenRouter rates before its credit fee.
| Model | Interactive | Batch API |
|---|---|---|
| GPT-6 Luna | $0.10 + $0.10 = $0.20 | $0.10 |
| GPT-6.1 Sol | $2.00 + $2.00 = $4.00 | $2.00 |
| Claude Opus 5.5 | $4.00 + $4.00 = $8.00 | $4.00 |
| GPT-6 Astra or Claude Fable 5.1 | $10.00 + $10.00 = $20.00 | $10.00 |
Two things follow. Claude Opus 5.5 in batch costs the same $4.00 as GPT-6.1 Sol interactive, so if the work can wait you can buy the top-ranked row at the mid-tier price. And GPT-6 Luna does the same volume for $0.20, which is where routine traffic belongs whichever vendor you pick for the hard work.
How the Batch API works: OpenRouter's Batch API went live in September. It bills most models at about 50% of the per-token price and returns results within 24 hours. Each batch uses one model and one endpoint shape, is text-only, and accepts chat completions, Responses and Anthropic Messages requests. xAI models are 20% off in batch rather than half, and web search and prompt caching bill at their usual rates. It suits evals, backfills and nightly summaries; it does not help when someone is waiting on the answer.
OpenRouter's credit fee sits on top of every figure above: 5.5% on the Standard plan, 8% on the self-serve Business plan. The $8.00 Opus 5.5 job costs about $8.44 in credits on Standard and $8.64 on Business.
Which Costs Less for Your Use Case
With flagship rates level, task mix matters less than it used to and output discipline matters more:
- Reading and analysing large documents → level on rate; pick on retrieval quality and context handling
- Generating long outputs → level on rate; the cheaper model is the one you edit less
- Short Q&A and chat at volume → the budget tier, not the flagship ($0.75–$1.00 per 1M input)
- Genuine bulk classification and extraction → a cheap high-throughput row such as DeepSeek V4 Flash at $0.04 input / $0.08 output per 1M
- Coding with a large codebase in context → check the context window as well as the rate
The single biggest saving on this page is no longer vendor choice. It is not sending flagship-tier traffic to a flagship model when a budget row would be accepted.
Practical Recommendation
If you use a subscription ($20/month): pick on features. The prices are identical and have been for a long time.
If you use the API: stop optimising the flagship choice. $2.00 versus $2.00 per 1M input is not where your bill lives. Your bill lives in the share of traffic that reaches a flagship model at all. Moving even a third of volume from $2.00/1M to $0.04/1M is a far larger cut than any vendor switch at the top tier.
Run both flagships on your own prompts, measure cost per accepted outcome including review time, and route everything that passes a cheap model to the cheap model.
The Budget Model Comparison
The budget tier is where the remaining price difference sits:
- OpenAI GPT Mini Latest: $0.75 input / $4.50 output per 1M tokens
- Anthropic Claude Haiku Latest: $1.00 input / $5.00 output per 1M tokens
- Google Gemini Flash Latest: $0.75 input / $3.75 output per 1M tokens
- DeepSeek V4 Flash: $0.04 input / $0.08 output per 1M tokens
Note the spread. The gap between the cheapest and dearest budget row is far wider than the gap between the two flagships, which is the opposite of how most teams spend their optimisation effort.
Key Takeaways
- →Subscriptions cost the same — $20/month for Plus or Pro, ~$25–30/user for Team plans
- →Flagship API rates have converged: $2.00/$10.00 per 1M versus $2.00/$10.00 per 1M
- →The “Claude is 40% cheaper on input” rule from 2024 no longer holds — do not route on it
- →The budget tier still spreads widely, from $0.04 to $1.00 per 1M input
- →Routing traffic away from flagship models saves far more than switching between them
- →September rows: Claude Opus 5.5 at $4/$20 is twice GPT-6.1 Sol's $2/$10, GPT-6 Astra and Claude Fable 5.1 sit at $10/$50, and the Batch API halves each for work that can wait 24 hours
Editorial context
Who is this for?
Developers, startups, and teams who want to reduce their AI API or subscription costs without sacrificing quality.
When NOT to use this
Users who need real-time data, image generation, or proprietary enterprise integrations may need more specialised tools.
Pricing insights
AI pricing varies widely — some models charge per token while others use flat subscriptions. Token-based APIs are usually cheaper for moderate usage, while subscriptions suit power users with high and consistent volume.
Alternatives to consider
Consider DeepSeek V4 Flash for cost-effective coding and writing, Gemini 3.8 Flash for fast tasks, or Claude Haiku 4.5 for lightweight structured work. Use the calculator to compare your specific usage.
Final verdict
The cheapest AI tool is the one that fits your exact workload. Use the cost calculator and decision engine on this site to find your optimal stack — most users can cut AI spend by 50% or more.
Frequently Asked Questions
Is ChatGPT or Claude cheaper?
For individuals they cost the same: ChatGPT Plus and Claude Pro are both $20/month. ChatGPT Go at $8 is the cheapest paid step; ChatGPT Business is $25 per seat; Claude Max starts at $100 for heavy users.
Which API is cheaper, OpenAI or Anthropic?
They are level at the high-mid tier: Claude Sonnet 5 and GPT-6.1 Sol both list $2 input / $10 output per 1M tokens. At the top they are not: Claude Opus 5.5 is $4 / $20, and GPT-6 Astra and Claude Fable 5.1 are both $10 / $50. The real saving is routing routine work to a budget row such as GPT-6 Luna ($0.10 / $0.50), GPT-5.4 mini ($0.75 / $4.50), Claude Haiku 4.5 ($1 / $5) or DeepSeek V4 Flash ($0.05 / $0.16).
Does Claude Pro include Claude Code?
Yes — Claude Pro ($20) and Claude Max (from $100) include Claude Code usage within plan limits. On ChatGPT's side, Codex access depends on the plan; see the OpenAI Codex pricing page.
Should I pay for both?
Rarely. Pick one $20 subscription on workflow fit, and use the API for anything automated, where model choice and caching matter more than vendor.
How much does Claude Opus 5.5 cost compared with GPT-6.1 Sol?
Claude Opus 5.5 is $4 input / $20 output per 1M tokens and GPT-6.1 Sol is $2 / $10, so Opus 5.5 is twice the rate. For 1M input plus 200K output tokens that is $8.00 on Opus 5.5 and $4.00 on GPT-6.1 Sol; through OpenRouter's Batch API the same job is $4.00 and $2.00.
Is the Batch API worth using for ChatGPT or Claude API work?
For anything that can wait up to 24 hours, yes. OpenRouter's Batch API bills most models at about half the per-token price: GPT-6 Luna $0.05 / $0.25, GPT-6.1 Sol $1 / $5, Claude Opus 5.5 $2 / $10, GPT-6 Astra and Claude Fable 5.1 $5 / $25 per 1M. Batches are text-only and use one model each, so they fit evals, backfills and nightly jobs, not live chat.
Related
Free courses · no sign-up
Still deciding? Learn the basics first, then come back to the prices.