OpenAI

GPT-5 — Benchmark Scores, Pricing & Performance Analysis

FLAGSHIPOpenAI
Chatbot Arena ELO
1390
Output Speed
100 tok/s
Input Cost
$1.3/1M
Output Cost
$10.0/1M
Context Window
400K
Max Output
128K
Knowledge Cutoff
Sept 2024
Accepts
text, image, file

GPT-5 by OpenAI demonstrates strong general intelligence, excellent coding ability, outstanding mathematical reasoning. View detailed benchmark data including scores across coding, math, reasoning, speed, and cost metrics.

Specifications last checked against the provider on . Benchmark scores carry their own source, linked beside each number.

General Benchmarks

MMLU-Pro
86.0%7th of 66

Massive Multitask Language Understanding professional benchmark

HuggingFace Leaderboard
Chatbot Arena ELO
139020th of 74

Human preference ELO from blind head-to-head votes

LMSYS / HuggingFace
IFEval
90.0%8th of 25

Strict instruction following accuracy on verifiable constraints

Google Research

Coding Benchmarks

Website Arena ELO
119454th of 108

Human preference ELO for building a working web page from a brief

Design Arena (via OpenRouter)
LiveCodeBench
78.0%5th of 66

Live competitive programming benchmark

livecodebench.github.io
HumanEval+
92.0%6th of 66

Code generation correctness with extended tests

Papers
SWE-bench Verified
74.9%12th of 67

Real-world software engineering task resolution

swebench.com
AA Coding Index
37.860th of 96

Artificial Analysis composite coding score

Artificial Analysis (via OpenRouter)
Design Arena ELO
117258th of 120

Human preference ELO for building things — websites, UI, games, charts, SVG

Design Arena (via OpenRouter)
BFCL
59.2%5th of 7

Berkeley Function Calling Leaderboard — tool use accuracy

Berkeley

Math Benchmarks

GSM8K
98.0%5th of 66

Grade school math word problems

Papers
MATH
94.6%9th of 66

Competition mathematics problem solving

Papers
AIME 2025
94.6%7th of 17

American Invitational Mathematics Examination — competition math

Papers

Reasoning Benchmarks

τ²-Bench Airline
68.4%53rd of 93

Multi-turn service agent making tool calls under strict policy constraints

OpenRouter (measured)
GPQA Diamond
86.0%28th of 140

Graduate-level science Q&A by domain experts

Papers
ARC-AGI
50.0%10th of 66

Abstraction and Reasoning Corpus for general intelligence

arcprize.org

Speed Benchmarks

Time to First Token
320 ms28th of 92

Latency before first token arrives

Aggregated
Output Speed
100 tok/s31st of 94

Tokens generated per second

Aggregated

Cost Benchmarks

Input Cost
$1.3306th of 413

Cost per 1M input tokens

OpenRouter API
Cached Input Cost
$0.1369th of 136

Cost per 1M cached input tokens — the price that actually applies to a long agent conversation

OpenRouter API
Output Cost
$10.0335th of 413

Cost per 1M output tokens

OpenRouter API

Context Benchmarks

Context Length
400K90th of 451

Maximum context window size

OpenRouter API

Available from 30 providers

The same model costs different amounts depending on who serves it. You can bring your own key for 4 of these — connect it here.

ProviderInput / 1MOutput / 1MCached inContextYour key
OpenCode Zen$1.07$8.5$0.107400K—
Poe$1.1$9$0.11400K—
302.AI$1.25$10—400K—
Abacus$1.25$10$0.125400K—
Azure$1.25$10$0.13400KSupported
Azure Cognitive Services$1.25$10$0.13400K—
Cloudflare AI Gateway$1.25$10$0.125128K—
Databricks$1.25$10$0.125400K—
DevPass (LLM Gateway)$1.25$10$0.125400K—
Eden AI$1.25$10$0.125400K—
FastRouter$1.25$10$0.125400K—
Impossibl$1.25$10$0.125400K—
Kilo Gateway$1.25$10$0.125400K—
LLM Gateway$1.25$10$0.125400K—
Merge Gateway$1.25$10$0.125400K—
NanoGPT$1.25$10$0.125400K—
NEAR AI Cloud$1.25$10$0.125400K—
Neon$1.25$10$0.125400K—
Ofox$1.25$10$0.13400K—
OpenAI$1.25$10$0.125400KSupported
OpenRouter$1.25$10$0.125400KSupported
OrcaRouter$1.25$10$0.125400K—
SAP AI Core$1.25$10$0.125400K—
Vercel AI Gateway$1.25$10$0.125400KSupported
ZenMux$1.25$10$0.12400K—
Cortecs$1.375$10.96$0.156400K—
Requesty$1.375$11$0.1375400K—
AnyAPI———400K—
Model Oracle AI———400K—
Snowflake Cortex———400K—

GPT-5 — Benchmark Scores Overview

Scores normalized to percentage scale for visual comparison. ELO scores mapped to 0-100 range (1100-1500).

GPT-5 — Frequently Asked Questions

How intelligent is GPT-5?

GPT-5 scores 1390 on the Chatbot Arena ELO rating, making it a high-performing AI model. This score is based on blind head-to-head human preference voting.

How much does GPT-5 cost?

GPT-5 costs $1.3 per 1M input tokens and $10.0 per 1M output tokens. This is mid-range pricing for its capability level.

How fast is GPT-5?

GPT-5 generates output at 100 tokens per second, which is moderate compared to other models. The time to first token is 320 ms.

How good is GPT-5 at coding?

GPT-5 achieves 74.9% on SWE-bench Verified, demonstrating excellent real-world software engineering capability. This benchmark tests the model's ability to resolve actual GitHub issues.

How good is GPT-5 at math and reasoning?

GPT-5 scores 94.6% on the MATH benchmark (competition-level mathematics). It also achieves 86.0% on GPQA Diamond, a graduate-level science reasoning benchmark.

What is the context window of GPT-5?

GPT-5 has a context window of 400K tokens. This determines how much text, conversation history, and code the model can process in a single request.

Who created GPT-5?

GPT-5 was created by OpenAI. It is classified as a flagship model in our catalogue.

Is GPT-5 open source?

No, GPT-5 is a proprietary model. It is available through OpenAI's API and compatible providers.