AI pricing in 2026: ChatGPT, Claude, Gemini, Grok and API costs compared
Last verified: — against each provider's official pricing page. We re-check monthly.
What every major AI assistant costs per month, what their APIs charge per million tokens, and what a real task costs once you count the retries. US prices before tax.
Key findings
- For the same task, the most expensive API model costs about 100× the cheapest.
- Three attempts at a task cost about 3.5× one attempt, not 3×: every retry resends the whole conversation.
- Standard plans cluster around $20 a month. ChatGPT, Gemini and Grok also sell a budget tier for $5–8, and power tiers now run from $100 to $500.
Subscription plans (per month)
Monthly price for one person. Higher tiers add usage, not just features.
| Assistant | Plans and monthly price | Source |
|---|---|---|
| ChatGPT OpenAI | FreeGo: $8 · may include adsPlus: $20Pro 100: $100Pro 200: $200Pro 500: $500 · includes Ultrafast | |
| Claude Anthropic | FreePro: $20 · $17/mo billed yearlyMax 5x: $100Max 20x: $200 | |
| Gemini Google | FreeGoogle AI Plus: $4.99Google AI Pro: $19.99Google AI Ultra (5x): $99.99Google AI Ultra (20x): $199.99 | |
| Grok xAI | FreePlus: $8Premium: $30Super: $100Ultra: $200 | |
| Microsoft Copilot Microsoft | FreeMicrosoft 365 Personal: $9.99Microsoft 365 Family: $12.99Microsoft 365 Premium: $19.99 | |
| Perplexity Perplexity | FreePro: $17 · per month, billed yearlyMax: $167 · per month, billed yearly | |
| Mistral Vibe Mistral | FreePro: $14.99Team (per user): $24.99 |
API prices per million tokens
Standard tier, short prompts. "Cost per 1,000 requests" assumes 1,500 input and 500 output tokens each — a typical short task.
Most capable
| Model | Input $/1M | Output $/1M | Cost per 1,000 requests |
|---|---|---|---|
| GPT-6 Astra · OpenAI | $10.00 | $50.00 | $40.00 |
| Claude Fable 5.1 · Anthropic | $10.00 | $50.00 | $40.00 |
| Claude Opus 5.5 · Anthropic | $4.00 | $20.00 | $16.00 |
| Gemini 4 Argon · Google Launch price. Rises to $4 / $20 on Dec 1, 2026. | $2.00 | $10.00 | $8.00 |
| Grok 4.7 · xAI | $2.00 | $6.00 | $6.00 |
| DeepSeek V4 Pro · DeepSeek Peak-hour price. Half price off-peak (outside 01–04 and 06–10 UTC on weekdays). | $1.32 | $3.96 | $3.96 |
| Mistral Large 4 · Mistral Sale price. List price $1.36 / $4.18. | $0.68 | $2.09 | $2.07 |
Mid-range
| Model | Input $/1M | Output $/1M | Cost per 1,000 requests |
|---|---|---|---|
| Gemini 3.1 Pro (preview) · Google | $2.00 | $12.00 | $9.00 |
| GPT-6.1 Sol · OpenAI | $2.00 | $10.00 | $8.00 |
| Claude Sonnet 5.5 · Anthropic | $2.00 | $10.00 | $8.00 |
| Mistral Medium 3.5 · Mistral | $1.50 | $7.50 | $6.00 |
| Grok 4.3 · xAI | $1.25 | $2.50 | $3.13 |
| Gemini 3.8 Flash · Google Rises to $1.50 / $7.50 on Jan 1, 2027. | $0.75 | $3.75 | $3.00 |
Small and cheap
| Model | Input $/1M | Output $/1M | Cost per 1,000 requests |
|---|---|---|---|
| Gemini 3.5 Flash-Lite · Google | $0.30 | $2.50 | $1.70 |
| DeepSeek V4.1 Flash · DeepSeek Peak-hour price. Half price off-peak (outside 01–04 and 06–10 UTC on weekdays). | $0.30 | $1.20 | $1.05 |
| Mistral Small 4 · Mistral | $0.15 | $0.60 | $0.53 |
| GPT-6 Luna · OpenAI | $0.10 | $0.50 | $0.40 |
| Claude Haiku 5.5 · Anthropic For prompts up to 100K tokens; $0.50 / $2.50 above. | $0.10 | $0.50 | $0.40 |
Calculator: what does a task really cost?
Pick a task, how many attempts it usually takes you, and how often you do it. The table shows the monthly API cost per model.
| Model | 1 attempt · Per month | 3 attempts · Per month | Extra cost of retries |
|---|---|---|---|
| GPT-6 Luna · OpenAI | $0.16 | $0.54 | +$0.39 |
| Claude Haiku 5.5 · Anthropic | $0.16 | $0.54 | +$0.39 |
| Mistral Small 4 · Mistral | $0.19 | $0.70 | +$0.51 |
| DeepSeek V4.1 Flash · DeepSeek | $0.39 | $1.40 | +$1.01 |
| Gemini 3.5 Flash-Lite · Google | $0.72 | $2.38 | +$1.66 |
| Mistral Large 4 · Mistral | $0.73 | $2.71 | +$1.98 |
| Grok 4.3 · xAI | $1.00 | $3.98 | +$2.98 |
| Gemini 3.8 Flash · Google | $1.16 | $4.07 | +$2.91 |
| DeepSeek V4 Pro · DeepSeek | $1.39 | $5.19 | +$3.80 |
| Grok 4.7 · xAI | $2.10 | $7.86 | +$5.76 |
| Mistral Medium 3.5 · Mistral | $2.33 | $8.15 | +$5.82 |
| Gemini 4 Argon · Google | $3.10 | $10.86 | +$7.76 |
| GPT-6.1 Sol · OpenAI | $3.10 | $10.86 | +$7.76 |
| Claude Sonnet 5.5 · Anthropic | $3.10 | $10.86 | +$7.76 |
| Gemini 3.1 Pro (preview) · Google | $3.60 | $12.36 | +$8.76 |
| Claude Opus 5.5 · Anthropic | $6.20 | $21.72 | +$15.52 |
| GPT-6 Astra · OpenAI | $15.50 | $54.30 | +$38.80 |
| Claude Fable 5.1 · Anthropic | $15.50 | $54.30 | +$38.80 |
Why retries cost more than you think
Chat APIs are stateless: on every new message, the whole conversation so far is sent again and billed again as input. So the second attempt pays for the first prompt, the first answer and your correction; the third pays for all of that again, plus more.
Example: drafting a report with Claude Sonnet 5.5. One good attempt costs $3.10 per 100 reports. Three attempts cost $10.86 — and that's before counting your time.
The cheapest way to use any model is to get a usable answer on the first try. That comes down to the prompt: who it is for, the context, the format and the length.
Optimize your prompt for free →Price changes to watch
- Gemini 4 Argon: Launch price. Rises to $4 / $20 on Dec 1, 2026.
- DeepSeek V4 Pro, DeepSeek V4.1 Flash: Peak-hour price. Half price off-peak (outside 01–04 and 06–10 UTC on weekdays).
- Mistral Large 4: Sale price. List price $1.36 / $4.18.
- Gemini 3.8 Flash: Rises to $1.50 / $7.50 on Jan 1, 2027.
- Claude Haiku 5.5: For prompts up to 100K tokens; $0.50 / $2.50 above.
How we check these prices
- Every figure comes from the provider's own pricing page or help center, linked below, on the date shown at the top.
- Prices are in US dollars before tax. App-store, regional and annual-billing prices can differ.
- API prices are the standard tier for short prompts. Batch processing, prompt caching and off-peak discounts can lower them; long prompts (usually above 200K tokens) cost more.
- Consumer plans change limits more often than prices. We list prices only; limits are described on each provider's page.
Frequently asked questions
Which AI subscription is the cheapest?
Excluding free plans, Google AI Plus ($4.99) is the cheapest paid tier we track, followed by ChatGPT Go and Grok Plus at $8. Most standard plans cost about $20 a month.
Is it cheaper to use the API than a subscription?
For light or occasional use, usually yes: with small models, a thousand short tasks can cost under a dollar. For heavy daily use of top models, a subscription is often cheaper and more predictable.
Why is output more expensive than input?
Generating tokens takes more computing than reading them, so providers usually charge 3 to 5 times more for output, and up to about 8 times on some small models. Asking for exactly the length you need is one of the simplest ways to cut cost.
How often do AI prices change?
Often. In 2026 several providers launched new tiers, ran launch discounts and scheduled increases. That's why every table here shows the date it was last verified.
More AI references
Sources
- OpenAI Help — What is ChatGPT Plus
- OpenAI — Introducing ChatGPT Go
- OpenAI Help — About ChatGPT Pro tiers
- OpenAI API pricing
- Claude pricing
- Claude Help — How much does the Max plan cost
- Google AI plans
- Gemini API pricing
- xAI pricing
- xAI API pricing
- Perplexity pricing
- Microsoft 365 Premium
- Mistral pricing
- Mistral API pricing
- DeepSeek API pricing