GPT-5.4 Mini
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.
GPT-5.4 Mini is multimodal AI model from OpenAI, priced at €0.001 per 1M input tokens with a 400K tokens context window.
About this model
0.7
Pricing
API Integration
Use our OpenAI-compatible API to integrate GPT-5.4 Mini into your application.
npm install railwailimport railwail from "railwail";
const rw = railwail("YOUR_API_KEY");
// Simple — just pass a string
const reply = await rw.run("gpt-5-4-mini", "Hello! What can you do?");
console.log(reply);
// With message history
const reply2 = await rw.run("gpt-5-4-mini", [
{ role: "system", content: "You are a helpful assistant." },
{ role: "user", content: "Explain quantum computing simply." },
]);
console.log(reply2);
// Full response with usage info
const res = await rw.chat("gpt-5-4-mini", [
{ role: "user", content: "Hello!" },
], { temperature: 0.7, max_tokens: 500 });
console.log(res.choices[0].message.content);
console.log(res.usage);Deep dive — OpenAI's GPT-5.4 Mini
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.
Visit OpenAI →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
- 400K tokens
- 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.
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.
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
Frequently asked questions
Related Models
View all MultimodalClaude Opus 4.7
Anthropic's April 2026 flagship. 87.6% on SWE-bench Verified, 3x higher image resolution, output self-verification, vision + reasoning.
Claude Sonnet 4.6
Anthropic's balanced mid-tier model from February 2026. Best price/performance for production workloads: 5x cheaper than Opus, near-flagship quality.
Depth Anything v2
Monocular depth-estimation model trained on 595k labeled and 62M unlabeled images. Strong zero-shot generalization in indoor and outdoor scenes.
GPT-5.4
OpenAI's unified flagship combining GPT and o-series reasoning into one model. 1M context, multimodal, top SWE-Bench Pro and OSWorld scores.
Start using GPT-5.4 Mini today
Get started with free credits. No credit card required. Access GPT-5.4 Mini and 100+ other models through a single API.