GPT-5.4

LLM Provider / Model · Production agentic coding, multi-step tool use, balanced cost/quality on long-context tasks, GPT-5.5 alternative at lower cost

At a glance

Input price	$3
Output price	$15
Cache price	$0.30
Price tier	Premium
Context window	1M+
Max output	64K
Context tier	500K+
Speed tier	Standard
Latency	medium
Knowledge cutoff	Oct 2025
Modality	Multimodal (vision)
Model ID	gpt-5-4
Provider	OpenAI
HumanEval	94%
MMLU	88%
SWE-Bench	74%
Benchmark	SWE-bench Pro: high
Released	Mar 2026
Hosting	Closed/API
Capabilities	Vision, Tool use, Streaming, Structured output, Function calling, Prompt caching, Agentic workflows

Vision, Tool use, Streaming, Structured output, Function calling, Prompt caching, Agentic workflows

Production agentic coding, multi-step tool use, balanced cost/quality on long-context tasks, GPT-5.5 alternative at lower cost

Kiro's spec mode is tuned for Claude models — GPT-5.4 may work but spec-driven flows are best with Claude.
Gemini CLI is locked to Google Gemini models — your LLM is GPT-5.4.
Claude Code only works with Claude models — your LLM is GPT-5.4.
Devin uses Cognition's proprietary model internally — your selected GPT-5.4 will be ignored.
OpenAI Codex (Cloud) runs GPT-5.1 Codex internally — your selected GPT-5.4 will be ignored.
Jules runs Gemini models internally — your selected GPT-5.4 will be ignored.
The Claude Code SDK is tuned for Claude models — GPT-5.4 may work via the model adapter but loses Claude-specific features (thinking, prompt caching, sub-agents).
Lovable uses a fixed internal model — your selected GPT-5.4 will be ignored.
Replit Agent 3 uses its own model selection internally — your selected GPT-5.4 will be ignored.

Build a full stack around GPT-5.4 — Flowpicker shows compatibility warnings before you commit.