GPT-5.4 Thinking

reasoning

OpenAI•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

Core Specs

1050K
Context Window
128K
Max Output
✓
Reasoning
✗
Open Source
Multimodal Support
textimage

Scenario Scores

Coding
#5 / 1492
Writing
#5 / 1488
Translation
#8 / 1486
Data Analysis
#1 / 1493
Conversation
#6 / 1485
Image Gen
—

User Feedback Highlights

Based on community feedback. Hover to see original reviews.

+ Native computer use + reasoning combined+ Strong on complex problem-solving− Reasoning tokens count toward output cost+ Adjustable reasoning effort levels+ Unified model (no separate Codex needed)− Higher latency at xhigh effort− GPT-5.2 Thinking users must migrate by June 2026
Sentiment:👍 52%😐 35%👎 13%

Pros & Cons

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

Reliability

SLA99.9%
Incidents (30d)0
与 GPT-5.4 共享基础设施，故障记录相同
View Status Page →

Pricing

Input (per 1M tokens)$2.50
Output (per 1M tokens)$15.00
Subscription$20/month
Updated on 2026-03-06

Compare with Others

GPT-5.4 Thinking vs GPT-5.4 →GPT-5.4 Thinking vs Claude Opus 4.6 →GPT-5.4 Thinking vs Claude Opus 4.5 →

Benchmarks

sweBenchPro57.7%
gdpval83%
osWorld75%
gpqaDiamond93.2%
arcAgi254.2%