How Much Does GPT-5 Cost?
Updated 12 July 2026 · first published 4 July 2026
GPT-5 is OpenAI's flagship reasoning model, released in 2026. It delivers substantially stronger performance on complex reasoning, coding, and multimodal tasks than GPT-4o - but at a premium price. This guide breaks down every cost dimension: API pricing, ChatGPT subscription tiers, how it compares to GPT-4o, and where cheaper alternatives like MiMO may be the smarter choice.
GPT-5 API pricing
OpenAI prices GPT-5 per million tokens, with separate rates for input, output, and cached input. Cached reads are significantly cheaper because the model reuses a previously computed prefix.
| Tier | Input $/1M | Output $/1M | Cache Read $/1M | Context Window |
|---|---|---|---|---|
| GPT-5.6-Sol (max) | $5.00 | $30.00 | $0.50 | 128K tokens |
| GPT-5.6-Terra (mid) | $2.50 | $15.00 | $0.25 | 128K tokens |
| GPT-5.6-Luna (budget) | $1.00 | $6.00 | $0.10 | 128K tokens |
| GPT-5.5 (flagship) | $5.00 | $30.00 | $0.50 | 128K tokens |
| GPT-5.4 (mid-tier) | $2.50 | $15.00 | $0.25 | 128K tokens |
OpenAI's new July 2026 generation (GPT-5.6) introduces three distinct tiers: Sol for frontier reasoning, Terra for balanced performance, and Luna for cost-conscious use cases. For comparison, older GPT-5.5 and 5.4 models remain available. GPT-5.6 and 5.5 at $5/$30 are significantly more capable than GPT-4o ($2.50/$10), but the effective cost depends on reasoning depth and cache efficiency.
ChatGPT subscription costs
If you use GPT-5 through the ChatGPT interface rather than the API, pricing is subscription-based:
| Plan | Monthly Price | GPT-5 Access | Key Limits |
|---|---|---|---|
| Free | $0 | GPT-5 mini only | Limited messages/day |
| Plus | $20/mo | GPT-5 with caps | ~80 GPT-5 messages/3hrs |
| Pro | $200/mo | Unlimited GPT-5 | No rate limits, voice, deep research |
| Team | $25/user/mo | GPT-5 with higher caps | Admin tools, SSO, higher limits |
| Enterprise | Custom | GPT-5 unlimited | SLA, dedicated support, data governance |
The Pro plan at $200/month is only cost-effective if you are a heavy daily user. For most developers and teams, the API with careful token management is cheaper than a Pro subscription.
GPT-5 vs GPT-4o: when does the upgrade make sense?
The price gap is real, so the question is whether GPT-5's quality advantage translates into cost savings elsewhere:
- Use GPT-5 when: the task requires multi-step reasoning, complex code generation, long-document analysis, or agentic workflows where a single high-quality pass saves multiple retry cycles.
- Use GPT-4o when: the task is straightforward - classification, extraction, summarization of short texts, simple chat, or high-volume batch processing where throughput matters more than per-request intelligence.
- Use GPT-5 mini or nano when: you need GPT-5-level instruction following but at lower cost, and can accept slightly reduced reasoning depth.
A practical rule: if your GPT-4o prompts include chain-of-thought instructions or multi-turn correction loops, GPT-5 may actually be cheaper because it does that reasoning natively in a single call.
How to reduce GPT-5 costs
Several strategies can cut your GPT-5 bill by 30–70% without sacrificing output quality:
- Leverage prompt caching. GPT-5's cache read rate ($2.50/1M) is 75% cheaper than the full input rate. Structure your prompts with a stable prefix (system instructions, context) and variable suffix (user query) to maximize cache hits.
- Use GPT-5 mini/nano for routing. Run a lightweight model first to classify intent, then escalate to full GPT-5 only for complex queries. This is the model routing pattern.
- Batch API for non-interactive work. OpenAI's batch API offers 50% discounts on asynchronous workloads - ideal for evaluation, enrichment, and report generation.
- Right-size context windows. Don't send 100K tokens of context when 5K will do. Every unnecessary input token costs 4× more on GPT-5 than GPT-4o.
- Consider alternatives. For many coding and reasoning tasks, models like MiMO, DeepSeek R1, or Claude Sonnet deliver comparable quality at a fraction of the cost. Use provider arbitrage to benchmark.
Hidden costs beyond the sticker price
The per-token rate is only part of the picture. Budget for these additional factors:
- Reasoning tokens. GPT-5 uses internal reasoning tokens for complex tasks. These are billed at the output rate ($30/1M) and can multiply your effective output volume by 2–5× on hard problems.
- Retry costs. If your prompts are ambiguous or your parsing is brittle, failed attempts still consume tokens. Robust prompt engineering pays for itself.
- Evaluation spend. Testing and benchmarking during development can easily exceed production spend if not tracked separately. See eval cost allocation.
Bottom line
GPT-5 is expensive but not uniformly so. At $10/$30 per million tokens, a typical API call with 1,000 input tokens and 500 output tokens costs about $0.025. For high-volume applications, the bill adds up fast - but smart caching, routing, and batching can reduce costs by half or more. Track every request, attribute spend to features, and benchmark alternatives regularly.
Related
- LLM Cost Benchmarks by Industry - modeled spend scenarios across customer support, finance, healthcare, and manufacturing.
- How do LLM providers charge? - the pricing models behind the numbers.
- Model routing - sending the right request to the right model.
- Provider arbitrage - finding the cheapest model that meets your quality bar.
- Prompt caching explained - how to actually get cache hits.
Want this applied to your own LLM spend? FinOps LLM runs a free audit of your AI costs and shows where the savings are. Book free audit →