GPT-5 API Pricing Comparison 2026: GPT-5.5 vs GPT-5.4 vs GPT-5.2 vs GPT-5.1
GPT-5 API Pricing Comparison: GPT-5.5 vs GPT-5.4 vs GPT-5.2 vs GPT-5.1
This guide compares GPT-5.5, GPT-5.4, GPT-5.2, and GPT-5.1 by price, context window, cached input, long-context rules, and practical workload fit.
GPT-5 API Pricing Comparison
| Model | EvoLink input / 1M | EvoLink output / 1M | EvoLink cached input / 1M | Reference input / 1M | Reference output / 1M | Context | Max output |
|---|---|---|---|---|---|---|---|
| GPT-5.5 | $4.00 | $24.00 | $0.40 | $5.00 | $30.00 | 1M | 128K |
| GPT-5.4 | $2.00 | $12.00 | $0.20 | $2.50 | $15.00 | 1.05M | 128K |
| GPT-5.2 | $1.75 | $14.00 | $0.175 | $1.75 | $14.00 | 400K | 128K |
| GPT-5.1 | $1.25 | $10.00 | $0.125 | $1.25 | $10.00 | 400K | 128K |
The short version:
| If your priority is... | Start with |
|---|---|
| Highest GPT route on EvoLink for demanding reasoning | GPT-5.5 |
| Long context plus lower flagship cost | GPT-5.4 |
| Production coding and reasoning at a balanced cost | GPT-5.2 |
| High-volume simple tasks | GPT-5.1 |
GPT-5.5 Pricing Details
GPT-5.5 is the premium GPT route on EvoLink. It is the model to test when your workload needs stronger reasoning, large-context analysis, or agent workflows where quality matters more than the lowest per-token price.
| Tier | Input | Output | Cached input |
|---|---|---|---|
| Reference pricing | $5.00 / 1M | $30.00 / 1M | $0.50 / 1M |
| EvoLink pricing | $4.00 / 1M | $24.00 / 1M | $0.40 / 1M |
| EvoLink long-context tier (>272K input) | $8.00 / 1M | $36.00 / 1M | Check live billing table |
GPT-5.4 Pricing Details
GPT-5.4 remains the lower-cost flagship route. It has a 1.05M context window, 128K max output, and a much lower output price than GPT-5.5.
| Tier | Input | Output | Cached input |
|---|---|---|---|
| Reference pricing | $2.50 / 1M | $15.00 / 1M | $0.25 / 1M |
| EvoLink pricing | $2.00 / 1M | $12.00 / 1M | $0.20 / 1M |
| EvoLink long-context tier (>272K input) | $4.00 / 1M | $18.00 / 1M | Check live billing table |
GPT-5.2 Pricing Details
GPT-5.2 is the production workhorse. It is cheaper than GPT-5.4 and GPT-5.5 on input and output, while still giving teams strong reasoning and coding performance for normal production contexts.
| Tier | Input | Output | Cached input |
|---|---|---|---|
| Reference pricing | $1.75 / 1M | $14.00 / 1M | $0.175 / 1M |
| EvoLink pricing | $1.75 / 1M | $14.00 / 1M | $0.175 / 1M |
GPT-5.1 Pricing Details
GPT-5.1 is the budget route for high-volume tasks that do not need the strongest reasoning.
| Tier | Input | Output | Cached input |
|---|---|---|---|
| Reference pricing | $1.25 / 1M | $10.00 / 1M | $0.125 / 1M |
| EvoLink pricing | $1.25 / 1M | $10.00 / 1M | $0.125 / 1M |
Cached Input Pricing: Why It Matters
All GPT-5 routes in this comparison support lower cached-input pricing. Cached input is useful when your application repeats the same system prompt, policy block, tool instructions, or retrieval scaffold across many requests.
| Model | EvoLink standard input | EvoLink cached input | Approx. cached-input savings |
|---|---|---|---|
| GPT-5.5 | $4.00 / 1M | $0.40 / 1M | 90% |
| GPT-5.4 | $2.00 / 1M | $0.20 / 1M | 90% |
| GPT-5.2 | $1.75 / 1M | $0.175 / 1M | 90% |
| GPT-5.1 | $1.25 / 1M | $0.125 / 1M | 90% |
To maximize cache hits, keep stable instructions identical across requests, put reusable context early, and avoid randomizing system-prompt wording.
Long-Context Pricing: Watch the 272K Input Threshold
For GPT-5.5 and GPT-5.4, very large prompts can move the full session into long-context pricing. On EvoLink, the long-context rule used for these routes is:
| Route | Trigger | Long-context input | Long-context output |
|---|---|---|---|
| GPT-5.5 | Input exceeds 272K tokens | $8.00 / 1M | $36.00 / 1M |
| GPT-5.4 | Input exceeds 272K tokens | $4.00 / 1M | $18.00 / 1M |
This matters for codebase analysis, legal review, research packs, and other workloads that regularly exceed 272K input tokens. If your prompt can be chunked without hurting answer quality, chunking may avoid long-context pricing.
GPT-5 vs Competitors: Price Comparison
Use this table as a rough price-positioning view, not a benchmark ranking. Model quality depends on the task, prompt, tool setup, latency target, and evaluation harness.
| Model | Input / output (per 1M) | Context | Best for |
|---|---|---|---|
| DeepSeek Chat | $0.27 / $1.10 | 64K | Budget tasks and high volume |
| Gemini 2.5 Flash | $0.30 / $2.50 | 1M | Fast long-context workloads |
| GPT-5.1 on EvoLink | $1.25 / $10.00 | 400K | Budget GPT route |
| Gemini 3.1 Pro | $2.00 / $12.00 | 1M | Multimodal and long-context workloads |
| GPT-5.2 on EvoLink | $1.75 / $14.00 | 400K | Production coding and reasoning |
| GPT-5.4 on EvoLink | $2.00 / $12.00 | 1.05M | Lower-cost GPT flagship route |
| Claude Sonnet 4.6 | $3.00 / $15.00 | 1M | Coding and agentic tasks |
| GPT-5.5 on EvoLink | $4.00 / $24.00 | 1M | Premium GPT route for hard reasoning |
| Claude Opus 4.6 | $5.00 / $25.00 | 1M | Research and complex reasoning |
EvoLink GPT-5 Pricing: How It Works
EvoLink is a unified API gateway that gives you one API key for GPT-5.5, GPT-5.4, GPT-5.2, GPT-5.1, Claude, Gemini, and other models.
- GPT-5.5: EvoLink listed pricing is $4.00 input / $24.00 output / $0.40 cached input per 1M tokens.
- GPT-5.4: EvoLink listed pricing is $2.00 input / $12.00 output / $0.20 cached input per 1M tokens.
- GPT-5.2: EvoLink page-level pricing is $1.75 input / $14.00 output / $0.175 cached input per 1M tokens.
- GPT-5.1: EvoLink page-level pricing is $1.25 input / $10.00 output / $0.125 cached input per 1M tokens.
- OpenAI-compatible integration: most teams only need to change the base URL, API key, and model value.
Cost Optimization Tips
1. Route by task complexity
Do not send every request to GPT-5.5. Use GPT-5.1 for simple extraction and summarization, GPT-5.2 for everyday production reasoning, GPT-5.4 for long-context work, and GPT-5.5 for the hardest tasks.
2. Maximize cached input
Stable system prompts, policy text, tool descriptions, and long instructions should stay byte-for-byte consistent where possible. Cached input can reduce the input side of the bill by roughly 90%.
3. Watch long-context thresholds
If a request exceeds 272K input tokens on GPT-5.5 or GPT-5.4, the full session can be billed at the long-context rate. Before sending a very large prompt, decide whether the job needs one full-context request or can be split into smaller passes.
4. Optimize for cost per successful task
The cheapest model per token is not always the cheapest model per completed workflow. Track retries, failure rate, manual review rate, latency, and the cost of tool calls when comparing GPT routes.
5. Use the newest route only where it changes the outcome
GPT-5.5 should earn its premium. If GPT-5.2 or GPT-5.4 solves the task reliably, keep that route as the default and reserve GPT-5.5 for escalations.
FAQ
How much does GPT-5 API cost?
On EvoLink, GPT-5.5 is listed at $4.00 input / $24.00 output per 1M tokens, GPT-5.4 at $2.00 / $12.00, GPT-5.2 at $1.75 / $14.00, and GPT-5.1 at $1.25 / $10.00. Cached input is priced lower across the GPT-5 family.
How much does GPT-5.5 API cost?
EvoLink lists GPT-5.5 at $4.00 per 1M input tokens, $24.00 per 1M output tokens, and $0.40 per 1M cached input tokens. For sessions above 272K input tokens, EvoLink's long-context GPT-5.5 tier is $8.00 input and $36.00 output per 1M tokens.
Which GPT-5 model is cheapest?
GPT-5.1 is the cheapest GPT-5 route in this comparison. It is best for high-volume tasks where the prompt is straightforward and the cost per request matters more than the strongest reasoning quality.
Is GPT-5.5 worth the higher price?
GPT-5.5 is worth testing when the task is difficult enough that better reasoning, fewer retries, or better long-context synthesis can offset the higher token price. If the task is routine, GPT-5.2 or GPT-5.4 will often be more cost-effective.
Should I use GPT-5.5 or GPT-5.4?
Start with GPT-5.4 if you need long context but want a lower output price. Test GPT-5.5 when the task is reasoning-heavy, agentic, or quality-sensitive enough to justify the premium.
What happens above 272K input tokens?
For GPT-5.5 and GPT-5.4 on EvoLink, inputs above 272K tokens can trigger long-context pricing for the full session. GPT-5.5 long-context pricing is $8.00 input / $36.00 output per 1M tokens; GPT-5.4 long-context pricing is $4.00 input / $18.00 output.
How much does a typical GPT-5 API call cost?
A call with 2,000 input tokens and 500 output tokens costs about $0.020 on GPT-5.5, $0.010 on GPT-5.4, $0.0105 on GPT-5.2, and $0.0075 on GPT-5.1 using EvoLink listed pricing before cached-input effects.
Does EvoLink use an OpenAI-compatible API?
Yes. EvoLink provides an OpenAI-compatible integration path for GPT models, so most teams can migrate by changing the base URL, API key, and model value rather than rewriting the application.
