Identity
- Status
- active
- Developer ID
gpt-5.4- Released
- Mar 5, 2026
- Version
- GPT-5.4
- License
- Unknown
OpenAIGPT-5.4
GPT-5.4 is an OpenAI reasoning model for coding, agentic tasks, and professional work.
- Token context
- 1,050K tokens
- Inputs
- Text, Image
- Outputs
- Text
- Released
- Mar 5, 2026
Providers & Pricing
Estimate workload cost →| Provider | Input / 1M | Output / 1M | Cache read / 1M | Cache write / 1M | Observed | Evidence |
|---|---|---|---|---|---|---|
OpenaiAvailable | $2.50 | $15.00 | $0.25 | — | Sep 3, 2026 | Provider-reported |
 OpenrouterAvailable | $2.50 | $15.00 | $0.25 | — | Oct 20, 2026 | Provider-reported |
Benchmarks
All benchmark results →| ARC-AGI-1verified-v1-302d8267c052 | 93.67 | — | — | GPT-5.4 (XHigh)unknown | Original sourcewinner eligible | Third-party benchmark |
| ARC-AGI-2verified-v2-bca68e12cc9a | 73.95 | — | — | GPT-5.4 (XHigh)unknown | Original sourcewinner eligible | Third-party benchmark |
| LiveBench2026-06-25-90f8d44da965 | 82.38 | 18,273 | 1,270 | gpt-5.4-xhighaverage per case | Recomputedwinner eligible | Third-party benchmark |
| LMArena Document Arenadocument-2026-09-13-a64ca3b12b6e | 1,471.00 | — | 33,331 | gpt-5.4; 95% CI [1464.96841782, 1477.04032901]; votes 33331; rank 14unknown | Original sourcewinner eligible | Third-party benchmark |
| LMArena Search Arenasearch-2026-10-13-a64ca3b12b6e | 1,199.08 | — | 126,730 | gpt-5.4-search; 95% CI [1193.60193056, 1204.56536227]; votes 126730; rank 21unknown | Original sourcewinner eligible | Third-party benchmark |
| LMArena Text Arenatext-2026-10-15-a64ca3b12b6e | 1,468.97 | — | 64,776 | gpt-5.4-high; 95% CI [1465.23802304, 1472.70880445]; votes 64776; rank 33unknown | Original sourcewinner eligible | Third-party benchmark |
| LMArena Text Arenatext-2026-10-15-a64ca3b12b6e | 1,452.22 | — | 67,870 | gpt-5.4; 95% CI [1448.56177849, 1455.87949508]; votes 67870; rank 48unknown | Original sourcewinner eligible | Third-party benchmark |
| LMArena Vision Arenavision-2026-10-09-a64ca3b12b6e | 1,303.15 | — | 29,881 | gpt-5.4-high; 95% CI [1297.39164127, 1308.90995855]; votes 29881; rank 18unknown | Original sourcewinner eligible | Third-party benchmark |
| LMArena Vision Arenavision-2026-10-09-a64ca3b12b6e | 1,292.11 | — | 22,832 | gpt-5.4; 95% CI [1285.63250026, 1298.59555072]; votes 22832; rank 27unknown | Original sourcewinner eligible | Third-party benchmark |
At a Glance
Model Facts
Capacity
- Token context
- 1,050K tokens
- Maximum output
- 128K tokens
- Total parameters
- Unknown
- Active parameters
- Unknown
- Knowledge cutoff
- Aug 31, 2025
Interface and Access
- Inputs
- Text, Image
- Outputs
- Text
- API available
- Yes
- Open weights
- No
- Self-hostable
- No
- Reasoning
- Yes
- Vision
- Yes
- Tool calling
- Yes
- Capabilities
- chat, generation, reasoning, structured outputs, tools
This page represents one developer model product. Dated API snapshots, serving endpoints, pricing, regions, quantizations, and service tiers attach as versioned aliases or provider details and never create duplicate public model records.
Version Lineage
PredecessorUnknown
SuccessorsUnknown
Aliasesgpt-5.4, gpt-5.4-2026-03-05
Explicit Unknowns
- Architecture
- Tokenizer
Primary Evidence
Sources and Observation Date
Comparable Peers
Related Model Comparisons
| A | Pair | B | Context |
|---|---|---|---|
GPT-5.4OpenAI | vs |  DeepSeek V4.1 FlashDeepSeek | cross-developer peersimage, text |
GPT-5.4OpenAI | vs | Claude Sonnet 5Anthropic | cross-developer peersimage, text |
GPT-5.4OpenAI | vs | cross-developer peersimage, text | |
GPT-5.4OpenAI | vs | Claude Opus 4.6Anthropic | cross-developer peersimage, text |
GPT-5.4OpenAI | vs | Claude Opus 4.8Anthropic | cross-developer peersimage, text |
GPT-5.4OpenAI | vs | Claude Opus 5Anthropic | cross-developer peersimage, text |
GPT-5.4OpenAI | vs | Claude Sonnet 4.6Anthropic | cross-developer peersimage, text |
GPT-5.4OpenAI | vs | GPT-4.1OpenAI | family variantsimage, text |
GPT-5.4OpenAI | vs | GPT-4.1 MiniOpenAI | family variantsimage, text |
GPT-5.4OpenAI | vs | GPT-5.3 CodexOpenAI | family variantsimage, text |
GPT-5.4OpenAI | vs | GPT-5.5OpenAI | family variantsimage, text |
GPT-5.4OpenAI | vs | Grok 4.3xAI | cross-developer peersimage, text |
Questions
GPT-5.4 FAQs
What is GPT-5.4?+
GPT-5.4 is an OpenAI reasoning model for coding, agentic tasks, and professional work. It is developed by OpenAI and its current sourced lifecycle status is active.
What inputs and outputs does GPT-5.4 support?+
GPT-5.4's current record lists Text and Image as input and Text as output.
How much context does GPT-5.4 support?+
The current record lists a 1,050K-token context window and a maximum output of 128K tokens.
Is GPT-5.4 available through an API?+
GPT-5.4 is marked as API-available. The directly sourced provider records currently include Openai and Openrouter.
How much does GPT-5.4 cost through an API?+
The lowest directly sourced prices currently attached to GPT-5.4 are $2.50 per million input tokens through Openai and $15.00 per million output tokens through Openai. Prices are provider-specific and should be checked against each cited observation date.
What capabilities and tasks does GPT-5.4 support?+
GPT-5.4's recorded capabilities are chat, generation, reasoning, structured outputs, and tools. Its supported tasks are image, reasoning, and text.
Can GPT-5.4 be self-hosted?+
GPT-5.4 is not marked as an open-weight model. Self-hosting is marked as unsupported, with no license recorded.
Does GPT-5.4 have related versions or aliases?+
No predecessor is recorded. No successor is recorded. Recorded aliases are gpt-5.4 and gpt-5.4-2026-03-05.
Which models can GPT-5.4 be compared with?+
The current compatible peer set includes DeepSeek V4.1 Flash, Claude Sonnet 5, Grok Build 0.1, Claude Opus 4.6, and Claude Opus 4.8. Compatibility requires shared sourced modalities and tasks; it does not imply that one model is better. Browse model comparisons →
How current is the Model Markets record for GPT-5.4?+
GPT-5.4 was last verified Sep 3, 2026 from OpenAI's primary source, “GPT-5.4 Model | OpenAI API.” The evidence is classified as official fact.