AI Model Cost Comparison — OpenAI, Anthropic & Google
Compare the cost of every major OpenAI, Anthropic, and Google Gemini model side by side. Enter your token usage to see exactly what each model would cost at your scale.
Prices sourced from OpenAI, Anthropic, and Google official pricing pages · Last updated: April 2026
Estimate costs at your usage volume
For this workload, Gemini 2.0 Flash is the cheapest at $7.50/mo — $1,342.50 (99%) less than Claude Opus 4.1.
Filter:
| Model | Input / 1M | Output / 1M | Est. monthly cost |
|---|---|---|---|
GoogleGemini 2.0 Flash Google's budget option — cheapest Gemini model for simple, high-volume tasks | $0.10 | $0.40 | |
OpenAIGPT-4o mini Fastest and cheapest GPT-4o variant — good for high-volume, latency-sensitive tasks | $0.15 | $0.60 | |
OpenAIGPT-5.4 nano Cheapest GPT-5.4-class model — built for simple, high-volume tasks | $0.20 | $1.25 | |
AnthropicClaude Haiku 3 Cheapest Claude model — great value for simple tasks that don't need the latest generation | $0.25 | $1.25 | |
OpenAIGPT-4.1 mini Lightweight GPT-4.1 — a solid default for most production workloads | $0.40 | $1.60 | |
GoogleGemini 2.5 Flash Fast and cost-efficient — the go-to Gemini model for production at scale | $0.30 | $2.50 | |
OpenAIGPT-5.4 mini Strongest mini model yet — strong on coding, computer use, and subagents | $0.75 | $4.50 | |
OpenAIo4-mini Compact reasoning model — cheaper than o3, still much stronger than GPT-4o on logic tasks | $1.10 | $4.40 | |
AnthropicClaude Haiku 4.5 Fastest and most cost-efficient Claude model — ideal for high-volume, latency-sensitive tasks | $1.00 | $5.00 | |
OpenAIGPT-4.1 Previous OpenAI flagship — still strong on coding, reasoning, and instruction following | $2.00 | $8.00 | |
OpenAIo3 OpenAI's reasoning model — slower and pricier, but significantly better at multi-step logic | $2.00 | $8.00 | |
GoogleGemini 2.5 Pro Previous Gemini flagship — 1M token context, widely deployed in production | $1.25 | $10.00 | |
OpenAIGPT-4o Multimodal model with vision — text, images, and audio in one API | $2.50 | $10.00 | |
GoogleGemini 3.1 ProPopular Google's latest flagship — 2M token context, strong on reasoning, code, and multimodal tasks | $2.00 | $12.00 | |
OpenAIGPT-5.4Popular OpenAI's most capable model for professional work — strongest reasoning and coding | $2.50 | $15.00 | |
AnthropicClaude Sonnet 4.6 Optimal balance of intelligence, cost, and speed — Anthropic's go-to for most production workloads | $3.00 | $15.00 | |
AnthropicClaude Sonnet 4 Previous Sonnet generation — still widely used in production | $3.00 | $15.00 | |
AnthropicClaude Opus 4.6Popular Anthropic's most intelligent model — best for complex agents, coding, and reasoning tasks | $5.00 | $25.00 | |
AnthropicClaude Opus 4.1 Legacy high-capability Opus model — available for workloads already built around it | $15.00 | $75.00 |
Sorted by estimated monthly cost (cheapest first) · Prices are for standard tier without caching or batch discounts · 1,000 req/day · 500 input + 500 output tokens/req
Full pricing reference
Raw prices per 1M tokens for quick reference, regardless of volume.
OpenAI
| Model | Input | Output |
|---|---|---|
| GPT-5.4 | $2.5/M | $15/M |
| GPT-5.4 mini | $0.75/M | $4.5/M |
| GPT-5.4 nano | $0.2/M | $1.25/M |
| GPT-4.1 | $2/M | $8/M |
| GPT-4.1 mini | $0.4/M | $1.6/M |
| GPT-4o | $2.5/M | $10/M |
| GPT-4o mini | $0.15/M | $0.6/M |
| o3 | $2/M | $8/M |
| o4-mini | $1.1/M | $4.4/M |
| Model | Input | Output |
|---|---|---|
| Gemini 3.1 Pro | $2/M | $12/M |
| Gemini 2.5 Pro | $1.25/M | $10/M |
| Gemini 2.5 Flash | $0.3/M | $2.5/M |
| Gemini 2.0 Flash | $0.1/M | $0.4/M |
Anthropic
| Model | Input | Output |
|---|---|---|
| Claude Opus 4.6 | $5/M | $25/M |
| Claude Sonnet 4.6 | $3/M | $15/M |
| Claude Haiku 4.5 | $1/M | $5/M |
| Claude Opus 4.1 | $15/M | $75/M |
| Claude Sonnet 4 | $3/M | $15/M |
| Claude Haiku 3 | $0.25/M | $1.25/M |
Once you're using more than one provider, the next problem is adding them up.
OpenAI's dashboard knows about OpenAI. Anthropic's knows about Anthropic. Neither knows about your customer list. PerUnit pulls usage from all three providers and joins it to your customers and features — one view, one bill, one answer to “what did this customer cost us this month?”
Get early access to PerUnitFrequently asked questions
- Is OpenAI cheaper than Anthropic?
- It depends on which model and what mix. At the flagship tier, GPT-4o ($2.50/M in, $10/M out) and Claude Sonnet 4.6 ($3.00/M in, $15/M out) are close, with OpenAI slightly cheaper on output. At the cheap tier the gap widens — GPT-4o mini ($0.15/M in) is meaningfully cheaper than Claude Haiku 4.5 ($1.00/M in). The use-case presets above will show you the answer at your actual volume mix.
- Which model gives the best performance per dollar?
- For high-volume but straightforward work — classification, extraction, summarisation, intent detection — GPT-4o mini, GPT-4.1 mini, and Gemini Flash all sit in roughly the same performance band at a fraction of flagship pricing. Routing the simple half of your traffic from a flagship to one of these typically cuts the bill 40–70% on those features without quality regression. The flagship is for the work that actually needs it.
- How does Google Gemini pricing compare to OpenAI and Anthropic?
- Gemini 2.5 Flash ($0.30/M in, $2.50/M out) sits between GPT-4o mini and GPT-4.1 mini on price. Gemini 2.5 Pro ($1.25/M in, $10/M out) is the cheapest of the three flagships on input, with GPT-4.1 next and Claude Sonnet most expensive. Gemini's bigger structural advantage is the 1M+ context window across tiers, vs 128K for OpenAI and 200K for Anthropic — relevant only if you actually use it.
- Do these providers offer volume discounts?
- Officially, no published tiers — but all three negotiate at enterprise volume. OpenAI's Batch API gives a flat 50% discount for async jobs (24h SLA). Prompt caching is available on OpenAI, Anthropic, and (implicitly on Flash) Google, and can drop repeated-context costs by 50–90%. The numbers above are standard tier; your real bill will likely be lower if any of these apply.
- Can I use all three providers in the same product?
- Yes — and many teams do. Gemini Flash for bulk summarisation, GPT-4o for multimodal, Claude Sonnet for long-context reasoning is a common split. The cost trade-off is that your spend is now split across three dashboards, none of which know about your customers. That split-brain attribution problem is the actual hard part of going multi-provider.
Get a monthly email when AI model pricing changes
One email a month. New model launches, price cuts, deprecations across OpenAI, Anthropic, and Google. No spam, unsubscribe any time.