GPT-5.4 API Guide: Pricing, Capabilities, and Migration from GPT-5.2
GPT-5.4 API Guide: Pricing, Capabilities, and Migration from GPT-5.2
GPT-5.4 is OpenAI’s latest frontier model, released March 5, 2026. Use gpt-5.4 in the API. It’s the first general-purpose model with native computer-use capabilities, supports 1M token context, and uses 47% fewer tokens than GPT-5.2 for complex tasks.
TL;DR for Developers
| What | Value |
|---|---|
| API Model Name | gpt-5.4 |
| Pro Version | gpt-5.4-pro |
| Input Price | $2.50 / 1M tokens |
| Output Price | $15 / 1M tokens |
| Cached Input | $0.25 / 1M tokens (90% savings) |
| Context Window | 1M tokens (experimental in Codex) |
| GPT-5.2 Deprecation | June 5, 2026 |
API Model Names
# Standard GPT-5.4
response = client.chat.completions.create(
model="gpt-5.4",
messages=[{"role": "user", "content": "..."}]
)
# Pro version for complex tasks
response = client.chat.completions.create(
model="gpt-5.4-pro",
messages=[{"role": "user", "content": "..."}]
)Pricing Comparison: GPT-5.4 vs GPT-5.2
| Model | Input | Cached Input | Output |
|---|---|---|---|
| gpt-5.4 | $2.50/M | $0.25/M | $15/M |
| gpt-5.4-pro | $30/M | — | $180/M |
| gpt-5.2 | $1.75/M | $0.175/M | $14/M |
| gpt-5.2-pro | $21/M | — | $168/M |
Cost change from GPT-5.2:
- Input: +43% ($1.75 → $2.50)
- Output: +7% ($14 → $15)
- BUT: Uses ~47% fewer tokens for tool-heavy workflows
Special pricing tiers:
- Batch/Flex: 50% of standard rate
- Priority processing: 2x standard rate
- Data residency/Regional: +10% for GPT-5.4 models
Key Capabilities (New in GPT-5.4)
1. Native Computer Use
First general-purpose model with built-in computer control. Operates via:
- Playwright code generation
- Mouse/keyboard commands from screenshots
- Customizable confirmation policies for safety
Benchmark: OSWorld-Verified
- GPT-5.4: 75.0% (exceeds human performance at 72.4%)
- GPT-5.2: 47.3%
2. 1M Token Context Window
Experimental support in Codex. Configure with:
model_context_window = 1000000
model_auto_compact_token_limit = ...⚠️ Requests exceeding 272K context count at 2x the normal rate.
3. Tool Search
For large tool ecosystems, GPT-5.4 can search through tools instead of loading all definitions upfront:
- Reduces token usage by 47% in MCP-heavy workflows
- Enables working with thousands of tools efficiently
4. Improved Reasoning Efficiency
Most token-efficient reasoning model yet:
- Fewer tokens to solve same problems vs GPT-5.2
- Faster speeds at same quality
/fastmode in Codex: 1.5x faster token velocity
Benchmark Improvements over GPT-5.2
| Benchmark | GPT-5.4 | GPT-5.2 | Improvement |
|---|---|---|---|
| GDPval (knowledge work) | 83.0% | 70.9% | +12.1% |
| OSWorld (computer use) | 75.0% | 47.3% | +27.7% |
| BrowseComp (web search) | 82.7% | 65.8% | +16.9% |
| SWE-Bench Pro (coding) | 57.7% | 55.6% | +2.1% |
| Toolathlon (tool use) | 54.6% | 45.7% | +8.9% |
| MMMU Pro (vision) | 81.2% | 79.5% | +1.7% |
Hallucination reduction:
- Individual claims: 33% less likely to be false
- Full responses: 18% fewer errors
Migration from GPT-5.2
Timeline
- Now: GPT-5.4 available as
gpt-5.4 - Now → June 5, 2026: GPT-5.2 in “Legacy Models” (ChatGPT)
- June 5, 2026: GPT-5.2 Thinking retired
Code Changes
# Before (GPT-5.2)
model = "gpt-5.2"
# After (GPT-5.4)
model = "gpt-5.4"Reasoning Effort Mapping
| GPT-5.2 | GPT-5.4 |
|---|---|
| none | none |
| low | low |
| medium | medium |
| high | high |
| heavy | xhigh |
New Parameters
model_context_window: Set context limit (up to 1M)model_auto_compact_token_limit: Auto-compact threshold- Tool search: Lightweight tool list + search capability
When to Use GPT-5.4 Pro
Use gpt-5.4-pro for:
- Maximum accuracy on complex tasks
- Long-horizon planning
- High-stakes outputs where cost is secondary
Pro benchmarks:
- BrowseComp: 89.3% (vs 82.7% standard)
- ARC-AGI-2: 83.3% (vs 73.3% standard)
- FrontierMath Tier 4: 38.0% (vs 27.1% standard)
ChatGPT Availability
| Plan | GPT-5.4 Thinking | GPT-5.4 Pro |
|---|---|---|
| Plus | ✅ | ❌ |
| Team | ✅ | ❌ |
| Pro | ✅ | ✅ |
| Enterprise | ✅ (admin enable) | ✅ |
| Edu | ✅ (admin enable) | ❌ |
FAQ
What’s the API model name for GPT-5.4?
Use gpt-5.4 for the standard model or gpt-5.4-pro for maximum performance.
How much more expensive is GPT-5.4 than GPT-5.2? Input is 43% more expensive ($1.75 → $2.50/M), output is 7% more ($14 → $15/M). However, GPT-5.4 uses ~47% fewer tokens for complex tool workflows, often reducing total cost.
When will GPT-5.2 be deprecated? GPT-5.2 Thinking will be retired on June 5, 2026. It’s available now under “Legacy Models” in ChatGPT.
Does GPT-5.4 support computer use?
Yes — it’s the first general-purpose model with native computer-use capabilities. Use the computer tool in the API.
What’s the context window? Standard: 272K tokens. Experimental 1M context available in Codex (counts at 2x rate above 272K).
How do I enable tool search? Provide a lightweight list of available tools; the model can search for full definitions when needed. See OpenAI’s updated documentation.
What’s the difference between GPT-5.4 and GPT-5.3-Codex? GPT-5.4 combines GPT-5.3-Codex’s coding strengths with knowledge work and computer-use capabilities. It matches or exceeds Codex on coding benchmarks while being faster.
Last verified: March 6, 2026
Sources:
- OpenAI: Introducing GPT-5.4 (March 5, 2026)
- OpenAI API Pricing
- TokenCost GPT-5.4 Analysis