GPT-5.4 Thinking
reasoningOpenAI•Released on 2026-03-05
GPT-5.4's reasoning variant with adjustable thinking depth. Replaces GPT-5.2 Thinking (deprecated June 2026). Supports four effort levels from 'low' to 'xhigh' for balancing speed vs reasoning depth. Available for Plus, Team, and Pro subscribers.
89
Overall Score
Voice of the community
sample 312“GPT-5.4 Thinking is rolling out gradually today in ChatGPT and Codex. Will replace GPT-5.2 Thinking, which is going away in three months.”
“I have talked a lot to 3.1 and compared it very directly to GPT 5.2 and Opus 4.6 and it feels like the most intelligent and most knowledgeable model when discussing difficult niche topics. It's just useless for agentic tasks.”
Core Specs
1050K
Context Window
128K
Max Output
ReasoningOpen Sourcetextimage
Scenario Scores
Pros & Cons
Sentiment52% +35% ·13% −
Pros
- +Adjustable reasoning effort levels
- +Strong on complex problem-solving
- +Unified model (no separate Codex needed)
- +Native computer use + reasoning combined
Cons
- −Higher latency at xhigh effort
- −Reasoning tokens count toward output cost
- −GPT-5.2 Thinking users must migrate by June 2026
Pricing
Input (per 1M tokens)$2.50
Output (per 1M tokens)$15.00
Subscription$20/month
Updated on 2026-03-06
Get Started
1Visit the provider's website
2Create an account
3Start using the model
Benchmarks
sweBenchPro57.7%
gdpval83%
osWorld75%
gpqaDiamond93.2%
arcAgi254.2%