What is GPT-5.4?
GPT-5.4 took over from GPT-5.2 as OpenAI's flagship for professional work and folded in the coding skills of GPT-5.3-Codex. It reads text and images, writes text, and works across a 1.05M-token window with up to 128K tokens of output; its knowledge stops at August 2025. It was OpenAI's first mainline model with built-in computer use and trained compaction.
It is still on the OpenAI API and Amazon Bedrock, but newer models now sit above it: GPT-5.5, the GPT-5.6 family and GPT-6. OpenAI has retired it from Codex for ChatGPT sign-ins and names GPT-6 Sol as the replacement.
What changed from GPT-5.2?
A much larger window, computer use and better agent work, for a slightly higher price.
- Context
- Grew from 400K to 1.05M tokens, with the same 128K-token output limit.
- Computer use
- Built in through the Responses API computer tool, which GPT-5.2 does not support.
- Agents
- Tool search and trained compaction arrived, and OpenAI reports gains in coding, tool use and long multi-step tasks.
- Price
- Up from $1.75 / $14 to $2.50 / $15 per million tokens, plus a surcharge past 272K input tokens.
Who is GPT-5.4 best for?
GPT-5.4 is best for developers who want a 1.05M-token OpenAI model with computer use for coding and professional work, at half the per-token price of GPT-5.5.
Computer use agents
It supports the computer use tool in the Responses API, so an agent can work through browser and desktop screens from screenshots.
Large codebases and document sets
A 1.05M-token context window with 128K tokens of output fits whole repositories or long reports in one request.
Half the price of GPT-5.5
At $2.50 / $15 per million tokens it costs half as much as GPT-5.5 with the same context size and tool set, and users say it holds up for everyday coding.
Hard questions with GPT-5.4 Pro
The Pro version spends more compute per answer for difficult analysis, at $30 / $180 per million tokens.
What's included with GPT-5.4?
Computer use, tool search, trained compaction, five reasoning levels, Batch, Flex and Fast mode, and a second home on Amazon Bedrock.
- Computer use
- Built into the Responses API: your code carries out the clicks and typing it returns, and OpenAI advises an isolated browser or VM.
- Tool search
- Loads only the tool definitions it needs at runtime, which saves tokens when an agent has many functions or MCP servers.
- Compaction
- Trained to compact long agent runs so they keep going without losing key context.
- Five reasoning levels
- Effort runs from none, the default, through low, medium and high to xhigh.
- Batch, Flex and Fast mode
- Half price for work that can wait, or $5 / $30 per million tokens for faster, steadier responses.
- Amazon Bedrock
- Also runs on Amazon Bedrock through an OpenAI-compatible Responses API, with a 1M-token window there.
How much does GPT-5.4 cost?
GPT-5.4 costs $2.50 per million input tokens and $15 per million output tokens on the OpenAI API, half that through Batch or Flex, and GPT-5.4 Pro costs $30 / $180.
- Cached input $0.13
- Batch returns results within 24 hours
- Flex answers slower and may turn requests away, unbilled
- Cached input $0.25
- 1.05M-token context window, 128K output
- Prompts over 272K tokens: $5 / $22.50
- Cached input $0.50
- Faster, steadier responses for user-facing apps
- More compute per answer; some requests take minutes
- Batch or Flex: $15 / $90
- Responses API only
When a request has more than 272K input tokens, the whole session is billed at twice the input rate and 1.5 times the output rate: $5 / $22.50 on Standard, $2.50 / $11.25 on Batch or Flex, and $60 / $270 for GPT-5.4 Pro. In Fast mode, prompts over 272K tokens run at Standard rates. Data residency endpoints add 10%. Tools such as web search are billed on top at their own rates. Full pricing on developers.openai.com. We checked these prices on .
How do readers rate GPT-5.4?
No one has rated GPT-5.4 yet. Be the first.
Sign in to rate. One rating per account.
GPT-5.4 pros and cons
Users praise its long, steady runs and close instruction following, but it gets expensive in heavy use and newer OpenAI models now beat it on price.
Pros
- Users report long unattended runs, from 30 to 90 minutes, that finish big task lists without stopping to ask whether to continue.
- Several developers say it follows specs and project instruction files closely, so they repeat themselves less.
- Developers use it to review code that Claude wrote and say it catches real problems.
- It costs half as much per token as GPT-5.5, and users cite the lower price as the reason they run it for everyday coding.
Cons
- Heavy use adds up: one developer spends $300 to $400 a month running it all day, and others say high effort, Fast mode and long context drain budgets and usage limits quickly.
- Prompts over 272K input tokens double the input rate and raise output by half for the whole session, so filling the 1.05M window is costly.
- Some Codex users found it falls short of GPT-5.5 on hard, rule-heavy work even at xhigh, where failed retries can cost more than using GPT-5.5.
- OpenAI's newer GPT-6 Sol costs less at $2 / $10 with the same window and newer knowledge, and has already replaced GPT-5.4 in Codex for ChatGPT sign-ins.
What are the best GPT-5.4 alternatives?
GPT-5.5 for harder work, GPT-5.4 mini for high-volume jobs, and Claude Sonnet 5 for a 1M-token window at a flat, lower price.
GPT-5.5PaidChoose it when GPT-5.4 still misses on hard, rule-heavy work; it keeps the 1.05M window and tool set, with knowledge to December 2025, at twice the price: $5 / $30.
GPT-5.4 mini & nanoPaidChoose it when volume matters more than depth; it keeps computer use and tool search at $0.75 / $4.50 per million tokens, with a 400K window.
Claude Sonnet 5FreeChoose it when you want a 1M-token window without a long-prompt surcharge; it costs $2 / $10 per million tokens at any length.
Holo4FreemiumChoose it when screen-driving agents need an Apache-licensed model you can self-host instead of OpenAI's native computer use.
Questions about GPT-5.4
What is GPT-5.4?
GPT-5.4 is OpenAI's model for coding and professional work, with a 1.05M-token context and built-in computer use, sold on the API at $2.50 / $15 per million tokens for developers building agents.
Who is GPT-5.4 for?
GPT-5.4 is best for developers who want a 1.05M-token OpenAI model with computer use for coding and professional work, at half the per-token price of GPT-5.5.
Is GPT-5.4 free?
No. GPT-5.4 is a paid tool, from $1.25 / $7.50 per million input / output tokens.
Does GPT-5.4 cost more for long prompts?
Yes. Once a GPT-5.4 request passes 272K input tokens, the whole session is billed at twice the input rate and 1.5 times the output rate: $5 and $22.50 per million tokens on Standard, or $2.50 and $11.25 on Batch or Flex.
What are the best GPT-5.4 alternatives?
The closest alternatives on Recaply are GPT-5.5, GPT-5.4 mini & nano, Claude Sonnet 5 and Holo4.
Where can I try GPT-5.4?
On its official website, developers.openai.com.
Can I still use GPT-5.4 in Codex?
Only with an API key. OpenAI retired GPT-5.4 from Codex for people who sign in with ChatGPT and suggests GPT-6 Sol instead. The OpenAI API and Codex signed in with your own API key still offer GPT-5.4, billed per token.
GPT-5.4 vs GPT-5.5: which should I use?
GPT-5.4 for everyday coding and agent work where cost matters, since it is half the price per token. GPT-5.5 costs $5 / $30, has newer knowledge, and some users say it handles hard, rule-heavy tasks that GPT-5.4 misses. Test both on your hardest prompts.
Should I switch from GPT-5.4 to GPT-6 Sol?
For new work, probably. GPT-6 Sol costs $2 / $10 per million tokens against $2.50 / $15, keeps the 1.05M-token window, and its knowledge runs to April 2026. OpenAI already names it as GPT-5.4's replacement in Codex. Test your prompts first, since behaviour can change.
Is GPT-5.4 available on Amazon Bedrock?
Yes. OpenAI lists GPT-5.4 among the models Amazon Bedrock runs through an OpenAI-compatible Responses API, with a 1M-token context window there. AWS handles access, regions and billing, and supported features vary by region.
Our verdict
GPT-5.4 is still a capable coding and agent model, but it is no longer the value pick in OpenAI's lineup.
Users rate its long, steady runs and close instruction following, yet GPT-6 Sol now costs less at $2 / $10 with the same 1.05M window and newer knowledge, and GPT-5.5 handles the hardest work.
- Skip it for
- New projects that can start on GPT-6 Sol, and workloads that often run past 272K input tokens.
- Check first
- Compare your bill and results against GPT-6 Sol before you commit more work to it.
GPT-6.1 Terra
MiniMax M3
Muse Spark 1.1
Qwen3.8-Flash
Kimi K3.1 Pro