OpenAI's reasoning-first flagship with configurable effort, 400K context, and strong real-world coding and agentic tool use.
Modality:
Text
Image
PDF
model ID
openai/gpt-5
Output Speed *
0.00
tok/s
Intelligence Index *
23
/ 100
Context Window *
400000
tokens
Input price
7.5
Anytoken
Output price
60
Anytoken
GPT-5: Configurable Reasoning for Coding and Long-Horizon Agents
GPT-5 is OpenAI's reasoning-first flagship, released August 2025 for coding, complex reasoning, and agentic task execution. It pairs a 400,000-token context window with configurable reasoning effort (minimal to high) and a verbosity control, letting one model span quick responses and deep multi-step problem solving. Its strongest evidence is real-world coding and tool orchestration: it scored 74.9% on SWE-bench Verified and 88% on Aider Polyglot with reasoning enabled. GPT-5 suits teams building coding agents and long chains of tool calls. OpenAI now positions it as its previous flagship, recommending later GPT-5.x models for new work.
Start building with the GPT-5 API on AnyAPI.ai
Performance
Where GPT-5 Earns Its Place: Coding and Tool Orchestration
GPT-5 is built for real-world software work and long chains of tool calls rather than single-shot chat. With reasoning enabled it reached 74.9% on SWE-bench Verified and 88% on Aider Polyglot, both measuring code changes against realistic repositories. That combination matters because production coding agents fail on multi-step edits and stale context, not toy prompts; a model that plans, calls tools, and revises across a 400K-token window resolves more issues end-to-end with less human intervention. The practical consequence is fewer supervised retries in agent loops, though enabling reasoning increases latency and token consumption on each request.
Benchmarks
GPT-5 Independent Benchmarks: Intelligence, Speed and Latency
Artificial Analysis rates GPT-5 (high) as above average in intelligence and faster than average, while noting it is somewhat verbose. On OpenAI's first-party API at high reasoning effort, independent measurement recorded roughly 95 output tokens per second and a time to first token near 91 seconds, reflecting the thinking phase reasoning models spend before answering. The verbosity and long time-to-first-token are the key selection signals: GPT-5 favors thoroughness over snappy responses, so it fits asynchronous agent and batch workloads better than latency-critical, interactive ones. Lowering reasoning effort trades intelligence for faster first tokens.
Output Speed
*
0.00
tok/s
Intelligence Index
*
23
/ 100
MMLU *
Broad world knowledge and problem-solving
87
%
GPQA *
PhD-level scientific reasoning across physics, biology, chemistry.
85
%
HLE *
Adherence to multi-step structured instructions.
28
%
LiveCodeBench *
Tool-calling reliability in long agentic loops.
85
%
Technical Specifications
What the model supports
GPT-5 accepts text and image input and returns text only; it is not a native audio, video, or image-generation model. It provides a 400,000-token context window and can produce up to 128,000 tokens in a single response — a large output ceiling that is separate from and much smaller than total context. The most production-relevant control is reasoning effort (minimal, low, medium, high) plus a verbosity parameter, letting teams tune the intelligence-versus-latency trade-off per request. Knowledge is current to September 30, 2024, so recent facts require retrieval or tool augmentation.
Verified Specifications —
GPT-5 (High)
*
Input modalities
Text
Image
PDF
output modalities
Text
Context window
400000
tokens
Maximum output tokens
128000
Reasoning
Yes
Knowledge cutoff
August 2025
Pricing (standard)
7.5
AnyTokens in
/
60
AnyTokens out
Quickstart
import requests
url = "https://api.anyapi.ai/v1/chat/completions"
payload = {
"model": "gpt-5",
"messages": [
{
"role": "user",
"content": [
{
"type": "text",
"text": "Text prompt"
},
{
"image_url": { "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" },
"type": "image_url"
}
]
}
]
}
headers = {
"Authorization": "Bearer AnyAPI_API_KEY",
"Content-Type": "application/json"
}
response = requests.post(url, json=payload, headers=headers)
print(response.json())import requests url = "https://api.anyapi.ai/v1/chat/completions" payload = { "model": "gpt-5", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Text prompt" }, { "image_url": { "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" }, "type": "image_url" } ] } ]
}
headers = { "Authorization": "Bearer AnyAPI_API_KEY", "Content-Type": "application/json"
} response = requests.post(url, json=payload, headers=headers) print(response.json())
View docs
Copy
Code is copied
const url = 'https://api.anyapi.ai/v1/chat/completions';
const options = {
method: 'POST',
headers: {Authorization: 'Bearer AnyAPI_API_KEY', 'Content-Type': 'application/json'},
body: '{"model":"gpt-5","messages":[{"role":"user","content":[{"type":"text","text":"Text prompt"},{"image_url":{"url":"https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"},"type":"image_url"}]}]}'
};
try {
const response = await fetch(url, options);
const data = await response.json();
console.log(data);
} catch (error) {
console.error(error);
}const url = 'https://api.anyapi.ai/v1/chat/completions';
const options = { method: 'POST', headers: {Authorization: 'Bearer AnyAPI_API_KEY', 'Content-Type': 'application/json'}, body: '{"model":"gpt-5","messages":[{"role":"user","content":[{"type":"text","text":"Text prompt"},{"image_url":{"url":"https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"},"type":"image_url"}]}]}'
}; try { const response = await fetch(url, options); const data = await response.json(); console.log(data);
} catch (error) { console.error(error);
}
View docs
Copy
Code is copied
curl --request POST \
--url https://api.anyapi.ai/v1/chat/completions \
--header 'Authorization: Bearer AnyAPI_API_KEY' \
--header 'Content-Type: application/json' \
--data '{
"model": "gpt-5",
"messages": [
{
"role": "user",
"content": [
{
"type": "text",
"text": "Text prompt"
},
{
"image_url": {
"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"
},
"type": "image_url"
}
]
}
]
}'curl --request POST \ --url https://api.anyapi.ai/v1/chat/completions \ --header 'Authorization: Bearer AnyAPI_API_KEY' \ --header 'Content-Type: application/json' \ --data '{ "model": "gpt-5", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Text prompt" }, { "image_url": { "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" }, "type": "image_url" } ] } ] }'
View docs
Copy
Code is copied
View docs
Limitations & Trade-offs
1
High time to first token. GPT-5 is a reasoning model, and independent measurement of the first-party API at high effort recorded a time to first token near 91 seconds because the model reasons before answering. For interactive, latency-sensitive UX such as live chat or autocomplete, this is disqualifying at high effort; lower reasoning effort helps, but latency-critical products are usually better served by a faster non-reasoning model or a lighter tier.
2
Verbosity and output cost. Artificial Analysis characterizes GPT-5 as somewhat verbose, and reasoning tokens plus long completions inflate output volume. Because output is the more expensive side of usage, high-effort runs and long generations become comparatively costly at scale. High-volume, cost-sensitive workloads should cap verbosity, lower reasoning effort, or route simpler requests to a cheaper model.
3
Knowledge cutoff of September 30, 2024. GPT-5 has no built-in awareness of events, libraries, or API changes after that date. For anything current — recent framework versions, news, or evolving documentation — you must supply context through retrieval or tools, or the model will answer from stale training data. Newer family members carry later cutoffs.
4
Superseded flagship status. OpenAI now labels GPT-5 as its previous model and recommends later GPT-5.x releases for new work, which score higher on current independent coding-agent benchmarks. This does not break existing deployments, but teams beginning new projects should weigh whether building on a superseded flagship is the right long-term foundation versus adopting the current recommendation.
Best-Fit Workloads
Where this model earns its place
01
Autonomous coding agents
GPT-5's 74.9% on SWE-bench Verified and 88% on Aider Polyglot with reasoning enabled make it a strong engine for agents that resolve GitHub issues, refactor across files, and ship patches end-to-end. Its support for long chains of tool calls lets an agent plan, run commands, inspect results, and revise. The 400K context holds substantial repository state. Budget for higher latency and token use per task at elevated reasoning effort.
02
Long-horizon tool-using assistants
Because GPT-5 executes extended sequences of tool calls and follows multi-step plans, it suits back-office assistants that query systems, transform data, and produce structured results across many steps. Structured outputs and function calling keep responses machine-parseable inside pipelines. This fits asynchronous or queued work where a first-token delay of tens of seconds is acceptable, rather than real-time chat.
03
Large-context analysis and review
The 400,000-token window supports codebase analysis, long technical documents, and multi-file review in a single request, and image input allows reasoning over diagrams and screenshots alongside text. Higher reasoning effort improves accuracy on dense material. Note the September 2024 knowledge cutoff means any current facts must come from the supplied context, not the model's memory.
04
Complex reasoning and STEM problem solving
GPT-5 scored 94.6% on AIME 2025 without tools and 84.2% on MMMU for multimodal understanding, indicating strong math and structured-reasoning ability. This fits technical analysis, quantitative problem solving, and evaluation tasks where correctness outweighs speed. Enable higher reasoning effort for the hardest problems and accept the added latency and token cost that thorough reasoning requires.
Integration
01
One API integration
02
Easy model switching
03
Flexible for production
04