GPT-5 Pricing Breakdown: Every Model, Every Tier, Every Cost
OpenAI's GPT-5 family now spans six models — from the ultra-cheap GPT-5 Nano at $0.05 per million input tokens to the powerhouse GPT-5.2 Pro at $21 per million input tokens. That's a 420x price difference within the same product family. Choosing the wrong tier could mean the difference between a $500/month API bill and a $50,000 one.
This guide breaks down every GPT-5 variant with real pricing data, cost calculations for common use cases, and clear recommendations for which model fits your workload. No hedging, no "it depends" — just numbers and decisions.
💡 Key Takeaway: Most teams should start with GPT-5 Mini ($0.25/$2 per million tokens) and only upgrade to GPT-5.2 when they hit a quality ceiling. The flagship models are 7-10x more expensive with marginal quality gains for typical tasks.
The Complete GPT-5 Model Lineup
Here's every model in the GPT-5 family with current API pricing:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Context Window | Max Output | Category |
|---|---|---|---|---|---|
| GPT-5 Nano | $0.05 | $0.40 | 128K | 16,384 | Budget |
| GPT-5 Mini | $0.25 | $2.00 | 500K | 65,536 | Efficient |
| GPT-5 | $1.25 | $10.00 | 1M | 131,072 | Flagship |
| GPT-5.1 | $1.25 | $10.00 | 1M | 131,072 | Flagship |
| GPT-5.2 | $1.75 | $14.00 | 1M | 131,072 | Flagship |
| GPT-5.2 Pro | $21.00 | $168.00 | 1M | 131,072 | Reasoning |
GPT-5 and GPT-5.1 share identical pricing at $1.25/$10, making GPT-5.1 the obvious choice between them — same cost, newer model, better performance. GPT-5.2 costs a 40% premium over GPT-5.1 on input and 40% on output.
[stat] 420x The price multiplier between GPT-5 Nano input ($0.05/M) and GPT-5.2 Pro input ($21/M)
Cost Per Request: What You Actually Pay
Per-million-token pricing is hard to reason about. Here's what each model costs for real tasks, assuming typical token counts:
Simple Tasks (500 input / 200 output tokens)
| Model | Cost Per Request | 10K Requests/Day | Monthly Cost |
|---|---|---|---|
| GPT-5 Nano | $0.000105 | $1.05 | $31.50 |
| GPT-5 Mini | $0.000525 | $5.25 | $157.50 |
| GPT-5 | $0.002625 | $26.25 | $787.50 |
| GPT-5.2 | $0.003675 | $36.75 | $1,102.50 |
| GPT-5.2 Pro | $0.044100 | $441.00 | $13,230.00 |
Complex Tasks (2,000 input / 1,000 output tokens)
| Model | Cost Per Request | 10K Requests/Day | Monthly Cost |
|---|---|---|---|
| GPT-5 Nano | $0.000500 | $5.00 | $150.00 |
| GPT-5 Mini | $0.002500 | $25.00 | $750.00 |
| GPT-5 | $0.012500 | $125.00 | $3,750.00 |
| GPT-5.2 | $0.017500 | $175.00 | $5,250.00 |
| GPT-5.2 Pro | $0.210000 | $2,100.00 | $63,000.00 |
📊 Quick Math: A SaaS product handling 10,000 complex requests per day pays $150/month with GPT-5 Nano or $63,000/month with GPT-5.2 Pro. That's not a rounding error — it's the difference between ramen profitability and needing Series A funding.
Long-Context Tasks (50,000 input / 2,000 output tokens)
Long-context workloads — document analysis, code review, RAG over large corpora — are where costs really diverge:
| Model | Cost Per Request | 1K Requests/Day | Monthly Cost |
|---|---|---|---|
| GPT-5 Nano | $0.003300 | $3.30 | $99.00 |
| GPT-5 Mini | $0.016500 | $16.50 | $495.00 |
| GPT-5 | $0.082500 | $82.50 | $2,475.00 |
| GPT-5.2 | $0.115500 | $115.50 | $3,465.00 |
| GPT-5.2 Pro | $1.386000 | $1,386.00 | $41,580.00 |
⚠️ Warning: Long-context requests are input-heavy, so the input price matters more than output price. GPT-5.2 Pro's $21/M input rate makes large document analysis extremely expensive. Consider chunking strategies or using GPT-5 Mini for initial passes.
GPT-5 Nano: The $0.05 Workhorse
GPT-5 Nano is OpenAI's cheapest model at $0.05 input / $0.40 output per million tokens. It's designed for high-volume, well-defined tasks where you need "good enough" quality at near-zero cost.
Best for:
- Classification and tagging
- Simple extraction (names, dates, amounts)
- Template-based text generation
- Routing requests to more expensive models
- Data validation and formatting
Not suitable for:
- Creative writing or nuanced analysis
- Complex reasoning chains
- Tasks requiring deep world knowledge
- Anything you'd show directly to a paying customer
GPT-5 Nano undercuts the older GPT-4o Mini by 3x on input, making it the best option for any task that doesn't require frontier intelligence. At scale, this adds up fast: a million daily classification requests costs roughly $1.50/day with Nano versus $4.50/day with GPT-4o Mini.
GPT-5 Mini: The Sweet Spot for Most Teams
At $0.25 input / $2.00 output per million tokens, GPT-5 Mini offers flagship-quality reasoning in a package that's 5x cheaper than GPT-5. For 80% of production workloads, this is the model you should be using.
Why GPT-5 Mini wins for most use cases:
- 500K context window covers nearly all document sizes
- 65K max output handles long-form generation
- Quality gap vs GPT-5 is marginal for structured tasks
- Cost is in the same ballpark as Gemini 2.5 Flash ($0.15/$0.60)
💡 Key Takeaway: GPT-5 Mini at $0.25/$2 sits in a competitive sweet spot. It's cheaper than Claude Haiku 4.5 ($1/$5) while matching it on most benchmarks. For teams committed to OpenAI's ecosystem, Mini is the default choice.
Cost comparison with competitors:
| Model | Input/M | Output/M | Provider |
|---|---|---|---|
| GPT-5 Mini | $0.25 | $2.00 | OpenAI |
| Gemini 2.5 Flash | $0.15 | $0.60 | |
| DeepSeek V3.2 | $0.28 | $0.42 | DeepSeek |
| Mistral Small 3.2 | $0.06 | $0.18 | Mistral |
| Claude Haiku 4.5 | $1.00 | $5.00 | Anthropic |
GPT-5 Mini's main disadvantage is output pricing — at $2/M, it's nearly 5x more expensive on output than DeepSeek V3.2 ($0.42/M). For generation-heavy tasks, DeepSeek or Mistral may be more economical. Use our cost calculator to compare for your specific token ratios.
GPT-5 vs GPT-5.1 vs GPT-5.2: Flagship Showdown
The three flagship GPT-5 models share 1M context windows and 131K max output, but differ in pricing and capability:
| Feature | GPT-5 | GPT-5.1 | GPT-5.2 |
|---|---|---|---|
| Input/M | $1.25 | $1.25 | $1.75 |
| Output/M | $10.00 | $10.00 | $14.00 |
| Release | Aug 2025 | Nov 2025 | Dec 2025 |
| Vision | ✅ | ✅ | ✅ |
| Audio | ❌ | ✅ | ✅ |
| Reasoning | Basic | Strong | Strongest |
The recommendation is straightforward: Use GPT-5.1 unless you specifically need GPT-5.2's enhanced coding and agentic capabilities. GPT-5.1 delivers 85-90% of GPT-5.2's performance at 71% of the input cost and 71% of the output cost. The original GPT-5 has no reason to exist in your stack anymore — GPT-5.1 is the same price with better performance.
✅ TL;DR: GPT-5.1 at $1.25/$10 is the best value flagship. Only upgrade to GPT-5.2 ($1.75/$14) if you're building coding tools or autonomous agents where the 10-15% quality improvement justifies a 40% cost increase.
GPT-5.2 Pro: When Money Is No Object
GPT-5.2 Pro is OpenAI's most expensive model at $21 input / $168 output per million tokens. It's a reasoning-specialized model that competes with o3-pro ($20/$80) but with significantly higher output costs.
[stat] $168/M GPT-5.2 Pro's output price — the most expensive per-token rate of any major API model
When GPT-5.2 Pro makes sense:
- Mathematical proofs and formal verification
- Complex multi-step research synthesis
- Legal or medical analysis where accuracy is critical
- Tasks where a single wrong answer costs more than $168
When it doesn't make sense (which is most of the time):
- Customer-facing chatbots (use GPT-5 Mini)
- Content generation (use GPT-5.1)
- Code completion (use GPT-5.2)
- Anything with more than 100 daily requests
Let's be concrete: if your application generates 500 tokens of output per request and you make 1,000 requests per day, GPT-5.2 Pro costs $84/day just in output tokens. GPT-5.2 handles the same load for $7/day. That's a 12x difference for a model that's marginally better on reasoning benchmarks.
⚠️ Warning: GPT-5.2 Pro's output pricing ($168/M) is more than double o3-pro's ($80/M). If you need OpenAI's best reasoning, compare both — o3-pro may deliver similar quality at lower output costs depending on your task profile.
OpenAI's Reasoning Models: o3 and o4-Mini
OpenAI also offers dedicated reasoning models that overlap with the GPT-5 family:
| Model | Input/M | Output/M | Best For |
|---|---|---|---|
| o4-mini | $1.10 | $4.40 | Efficient reasoning |
| o3 | $2.00 | $8.00 | General reasoning |
| o3-pro | $20.00 | $80.00 | Maximum reasoning |
| o1 | $15.00 | $60.00 | Legacy reasoning |
How to choose between GPT-5.2 and o3:
- GPT-5.2 ($1.75/$14) is better for general tasks with some reasoning
- o3 ($2/$8) is better when reasoning is the primary requirement — cheaper output
- o4-mini ($1.10/$4.40) is the budget reasoning pick — cheaper than GPT-5 Mini on output
For reasoning-heavy workloads, o4-mini at $1.10/$4.40 actually undercuts GPT-5 Mini ($0.25/$2) when output tokens exceed input tokens by 2:1 or more. Check our token calculator to model your specific ratio.
Cost Optimization Strategies for GPT-5
1. Use Model Routing
Don't send every request to the same model. Build a routing layer:
- Simple queries → GPT-5 Nano ($0.05/$0.40)
- Standard tasks → GPT-5 Mini ($0.25/$2.00), or GPT-5.4 mini when you need stronger coding and tool use
- Complex analysis → GPT-5.1 ($1.25/$10.00)
- Critical reasoning → GPT-5.2 Pro ($21/$168) — sparingly
A well-designed router can reduce costs by 60-80% compared to sending everything to a flagship model. Read our guide on AI API cost optimization strategies for implementation details.
2. Leverage the Batch API
OpenAI's Batch API offers 50% off on all models for non-time-sensitive requests. That brings GPT-5.2 down to roughly $0.875/$7 per million tokens — cheaper than standard GPT-5 pricing.
| Model | Standard Input/M | Batch Input/M | Savings |
|---|---|---|---|
| GPT-5 Nano | $0.05 | $0.025 | 50% |
| GPT-5 Mini | $0.25 | $0.125 | 50% |
| GPT-5.2 | $1.75 | $0.875 | 50% |
If your workload can tolerate 24-hour turnaround, the Batch API should be your default. See our OpenAI Batch API guide for setup instructions.
📊 Quick Math: An analytics pipeline processing 1M tokens of input daily through GPT-5.2 saves $0.875/day or $26.25/month per million tokens just by switching to batch. At 100M tokens/day, that's $2,625/month saved.
3. Prompt Compression and Caching
OpenAI's prompt caching automatically discounts repeated prefixes. For applications with system prompts or shared context:
- Cache hits cost 50% less on input tokens
- Long system prompts (1,000+ tokens) benefit most
- Combine with GPT-5 Mini for maximum savings
4. Consider Alternatives for Specific Tasks
OpenAI isn't always the cheapest option. For specific workloads:
- High-volume classification: Mistral Small 3.2 at $0.06/$0.18 beats GPT-5 Nano
- Long-form generation: DeepSeek V3.2 at $0.28/$0.42 has dirt-cheap output
- Balanced performance: Gemini 2.5 Flash at $0.15/$0.60 undercuts GPT-5 Mini
See our cheapest AI APIs guide for the full comparison.
Real-World Cost Scenarios
Startup SaaS (10K users, 50 requests/user/month)
- 500K requests/month, 800 avg input tokens, 400 avg output tokens
- GPT-5 Nano: $45/month
- GPT-5 Mini: $225/month
- GPT-5.2: $1,575/month
Recommendation: Start with GPT-5 Mini. At $225/month, it's a rounding error on your cloud bill and delivers quality users will notice.
Enterprise Analytics (100K documents/day)
- 100K requests/day, 10K avg input tokens, 1K avg output tokens
- GPT-5 Mini: $850/day → $25,500/month
- GPT-5.2: $5,650/day → $169,500/month
- GPT-5 Mini + Batch API: $425/day → $12,750/month
Recommendation: GPT-5 Mini with Batch API. You cut costs in half with no quality loss on batch-compatible workloads.
AI Agent Platform (Autonomous agents, 500 actions/agent/day)
- 1,000 agents, 500K requests/day, 2K input / 1K output tokens avg
- GPT-5.1: $62,500/day → $1,875,000/month
- GPT-5 Mini: $12,500/day → $375,000/month
- Mixed (Mini for planning, Nano for execution): ~$6,000/day → $180,000/month
Recommendation: Mixed routing is essential at this scale. Use our guide on AI agent costs for detailed architecture patterns.
✅ TL;DR: For most teams, GPT-5 Mini + Batch API + model routing delivers 90% of flagship quality at 10-20% of flagship cost. Only go flagship when quality demonstrably suffers on Mini.
Frequently asked questions
How much does GPT-5 cost per request?
A typical GPT-5 request with 1,000 input tokens and 500 output tokens costs $0.00625. GPT-5 Mini handles the same request for $0.00125, and GPT-5 Nano for $0.00025. Use our calculator to estimate costs for your specific token counts.
Which GPT-5 model should I use?
GPT-5 Mini ($0.25/$2 per million tokens) is the right choice for 80% of production workloads. Upgrade to GPT-5.1 ($1.25/$10) only if Mini's quality is measurably insufficient for your task. Use GPT-5 Nano ($0.05/$0.40) for classification, routing, and simple extraction tasks.
Is GPT-5.2 Pro worth the price?
For most applications, no. GPT-5.2 Pro costs $21/$168 per million tokens — 12x more than GPT-5.2 on output. It's justified only for high-stakes reasoning tasks where accuracy directly impacts revenue or safety: legal analysis, financial modeling, medical research. For everything else, GPT-5.2 or o3 delivers comparable results at a fraction of the cost.
How does GPT-5 pricing compare to Claude and Gemini?
GPT-5 Mini ($0.25/$2) is cheaper than Claude Haiku 4.5 ($1/$5) but more expensive than Gemini 2.5 Flash ($0.15/$0.60) on both input and output. At the flagship tier, GPT-5.2 ($1.75/$14) costs more than Claude Sonnet 4.6 ($3/$15) on input but less on output. The best value depends on your input-to-output ratio — check our cross-provider comparison.
Does OpenAI offer volume discounts on GPT-5?
OpenAI doesn't publish volume discounts, but the Batch API effectively gives 50% off for non-real-time workloads. Enterprise customers on committed spend agreements may negotiate custom rates. The Batch API is available to all developers and should be your first optimization lever.
Start Calculating Your GPT-5 Costs
The GPT-5 family offers a model for every budget, but choosing the right one requires understanding your actual usage patterns. Plug your numbers into our AI Cost Calculator to get precise monthly estimates across all GPT-5 models — and compare them against Claude, Gemini, Mistral, and DeepSeek in seconds.
Related reading: