OpenAI's efficient mid-tier model. 2x faster than its predecessor, 400k context, approaches GPT-5.4 quality on SWE-Bench Pro at a fraction of the cost.
Billed by the tokens actually used; the unused part of the reservation is refunded.
New here?
10 free credits (US$0.10) when you sign up with Google
Usable 24 hours after sign-up, up to 5 runs per day and at most 2 credits per run. Other sign-in methods start without credits. Enough for 17 runs of this model.
02
About GPT-5.4 Mini
TL;DRAs of September 23, 2026
GPT-5.4 Mini is a model by OpenAI in the Multimodal category. On Railwail, GPT-5.4 Mini costs US$0.90 per 1M input tokens and US$5.40 per 1M output tokens. The context window holds 400,000 tokens, and one response can be up to 128,000 tokens long.
Released March 17, 2026, GPT-5.4 mini brings the strengths of GPT-5.4 to a smaller, faster model designed for the subagent era. 400K context, vision input, integrated tool use, and 2x faster latency than GPT-5 mini. Significant gains on coding, reasoning, multimodal understanding and tool use; approaches the full GPT-5.4 on SWE-Bench Pro and OSWorld-Verified. Recommended for subagent workflows, customer-facing chat, coding assistants and high-volume API workloads.
Background
About OpenAI
Founded 2015 · San Francisco, USA
OpenAI was founded in December 2015 as a non-profit AI research organisation and transitioned to a capped-profit structure in 2019. The GPT lineage spans GPT-1 (2018) through GPT-5 (mid-2025) and the GPT-5.x family (2025-2026) which unified the o-series reasoning models with the general-purpose GPT line. GPT-5.4 mini and nano were announced on March 17, 2026 as the small-model tier of the GPT-5.4 generation. OpenAI is backed by Microsoft, Khosla, Andreessen Horowitz, Thrive Capital and Sequoia, with total funding above $60 billion and a 2026 valuation above $300 billion.
Unified Transformer (mid-tier, with integrated 'Thinking' tier)
GPT-5.4 mini was announced March 17, 2026 alongside GPT-5.4 nano as the small-model tier of the GPT-5.4 generation. It is a smaller variant of the unified GPT-5.4 architecture, retaining native text + image input, an integrated 'Thinking' tier for reasoning on demand, and the full tool-use API, while running more than 2x faster than GPT-5 mini at significantly lower cost. Pretraining used a similar multi-trillion-token mixture as GPT-5.4 with heavier distillation pressure from larger teacher models. Post-training included supervised fine-tuning, RLHF and reinforcement learning against verifiable rewards on coding, reasoning and tool-use trajectories. On evaluations such as SWE-Bench Pro and OSWorld-Verified, GPT-5.4 mini approaches the performance of full GPT-5.4 while costing roughly one-third as much.
Parameters
Undisclosed (estimated tens of billions of parameters, likely sparse MoE)
Context
400,000 tokens
Capabilities
2x faster than GPT-5 mini at lower cost
Approaches full GPT-5.4 on SWE-Bench Pro and OSWorld-Verified
400K token context window
Native multimodal input: text and images
Integrated 'Thinking' tier activates for harder reasoning
Native tool use, function calling and parallel tool calls
Designed for the subagent era: works well under an orchestrator
Strong coding, classification and extraction performance
Available in ChatGPT, Codex CLI and the OpenAI API
Regional processing endpoints available with 10% uplift
Best for: subagent workflows, customer-facing chat, coding assistants, high-volume API workloads.
Training & license
Pretrained on a multi-trillion-token mixture of web text, code, scientific papers and licensed data; heavy distillation from larger GPT-5.4 teacher models. Post-training uses supervised fine-tuning, RLHF and RL against verifiable rewards. Knowledge cutoff approximately late 2025.
License: Proprietary commercial license via OpenAI API and Azure OpenAI.
Safety testing: Evaluated under OpenAI's Preparedness Framework with internal and external red-teaming and capability evaluations.
Known limitations
Below full GPT-5.4 on the hardest agentic and reasoning benchmarks
Smaller context window than GPT-5.4 (400K vs 1.05M)
No native audio or video input
Knowledge cutoff in late 2025
Thinking mode adds latency and token cost when enabled
curl https://railwail.com/api/v1/chat/completions \
-H "Authorization: Bearer $RAILWAIL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5-4-mini",
"messages": [
{
"role": "user",
"content": "Explain what a vector database is in two sentences."
}
],
"max_tokens": 1024
}'
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["RAILWAIL_API_KEY"],
base_url="https://railwail.com/api/v1",
)
completion = client.chat.completions.create(
model="gpt-5-4-mini",
messages=[
{
"role": "user",
"content": "Explain what a vector database is in two sentences.",
},
],
max_tokens=1024,
)
print(completion.choices[0].message.content)
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.RAILWAIL_API_KEY,
baseURL: "https://railwail.com/api/v1",
});
const completion = await client.chat.completions.create({
model: "gpt-5-4-mini",
messages: [
{
role: "user",
content: "Explain what a vector database is in two sentences."
}
],
max_tokens: 1024
});
console.log(completion.choices[0].message.content);
// npm install railwail
import railwail from "railwail";
const rw = railwail(process.env.RAILWAIL_API_KEY);
const res = await rw.chat("gpt-5-4-mini", [
{ role: "user", content: "Explain what a vector database is in two sentences." },
], { max_tokens: 1024 });
console.log(res.choices[0].message.content);
GPT-5.4 Mini is a model by OpenAI in the Multimodal category. On Railwail you can call it with an API key through the Railwail API.
How much does GPT-5.4 Mini cost on Railwail?
On Railwail, GPT-5.4 Mini costs US$0.90 per 1M input tokens and US$5.40 per 1M output tokens. You are charged for what each request actually uses. Usage is paid from prepaid credits; 1 credit equals US$0.01.
What is the context window of GPT-5.4 Mini?
The context window of GPT-5.4 Mini holds 400,000 tokens. One response can be up to 128,000 tokens long.
How fast is GPT-5.4 Mini?
There are not enough measured runs of GPT-5.4 Mini on Railwail yet to state a run time. It depends on the input, the settings and the load at the provider.
Is GPT-5.4 Mini better than BLIP?
That depends on the task. GPT-5.4 Mini (OpenAI) and BLIP (Salesforce) are both models in the Multimodal category. The comparison page shows their prices and specifications side by side.
Yes. GPT-5.4 Mini accepts images as input in addition to text.
How do I use GPT-5.4 Mini through the API?
Create a Railwail API key and send your request with the model ID gpt-5-4-mini. Code examples for curl, Python and JavaScript are in the API section of this page.
