GPT-5.4
OpenAI's March 2026 model for coding and professional work, since superseded by GPT-5.5 and GPT-6
OpenAI has replaced this model with GPT-6.1 Sol.
Context window
1,050,000 tokens (128,000 max output)
Input / 1M tokens
$2.50
Output / 1M tokens
$15.00
Provider
OpenAI
Cached input $0.25 per 1M. Prompts exceeding 272K input tokens are billed at 2x input and 1.5x output for the full session. A gpt-5.4-pro variant is priced at $30 input / $180 output per 1M. · Data verified 2026-11-08
GPT-5.4 is a GPT-5-generation frontier model in the OpenAI API, with a snapshot dated March 5, 2026 (gpt-5.4-2026-03-05). It offers a 1,050,000-token context window with up to 128,000 output tokens, accepts text and image input, and has a knowledge cutoff of August 31, 2025. OpenAI describes it as a more affordable model for coding and professional work than GPT-5.5 ($2.50 input / $15 output per 1M). Newer models now undercut it: GPT-6.1 Sol costs $2/$10 and GPT-6.1 Terra $1/$5. A gpt-5.4-pro variant offers extended reasoning at higher cost. It stays in the API with no announced shutdown date.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Available in the OpenAI API via model id 'gpt-5.4' (snapshot 'gpt-5.4-2026-03-05') on the Responses, Chat Completions and Batch endpoints, plus a gpt-5.4-pro variant, and on Amazon Bedrock (openai.gpt-5.4). Still available, though OpenAI has released newer GPT-5.5, GPT-5.6 and GPT-6 models.
Strengths
- ✓Large 1.05M-token context window
- ✓Lower price than GPT-5.5 ($2.50/$15 vs $5/$30)
- ✓Solid general coding, reasoning, and tool use
- ✓Multimodal text + image input
Best for developers who...
When to choose it (and when not to)
Reach for GPT-5.4 when...
- →Existing GPT-5.4-based systems where behavior is already tuned and validated
- →Large-context tasks on an established GPT-5.4 integration
- →You need a pinned, dated snapshot for reproducible behavior
Look elsewhere if...
- ✕New projects: GPT-6.1 Sol is cheaper ($2/$10) and stronger
- ✕Tasks needing the strongest available reasoning/coding (use GPT-6 Astra or GPT-6.1 Sol)
- ✕Budgets sensitive to the >272K-token long-context surcharge
How to use it
- ›Reasoning effort defaults to none; set low through xhigh when a task needs more thinking
- ›Stay under 272K input tokens to avoid the long-context price multiplier
- ›Use cached input ($0.25/1M) for repeated large prefixes
- ›Pin the snapshot 'gpt-5.4-2026-03-05' for reproducible behavior
Quickstart
Pythonfrom openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="gpt-5.4",
input="Explain this stack trace and propose a fix.",
)
print(response.output_text)Snapshot 'gpt-5.4-2026-03-05' pins behavior. gpt-5.4-pro is available for deeper reasoning at higher cost.
API model id: gpt-5.4
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| SWE-bench Pro | 57.7% | Coding capability |
| OSWorld | 75.0% | Computer use; exceeds human baseline of 72.4% |
| GDPval | 83.0% | Professional knowledge work across 44 occupations |
| MMMU-Pro | 81.2% | Visual understanding and reasoning |
Compare GPT-5.4
Compare GPT-5.4 with any other model
Build a comparison →All model comparisons →