이 모델의 강점은 무엇인가요?

Advanced agentic programming capabilities Excellent computer operation functions Vast one-million-token context

이 모델의 약점은 무엇인가요?

Lack of transparency from closed-source API costs tied to usage Limited public licensing model

어떤 용도에 가장 적합한가요?

Developing autonomous AI agents Automating PC operation workflows Knowledge work on large documents

모델 목록으로

OpenAI독점

GPT-5.5

Name: GPT-5.5
Price: 5 USD
Author: OpenAI

GPT-5.5 is a flagship reasoning model developed by OpenAI. It specializes in agent programming, computer operation, and knowledge work, and supports a vast context window of 1 million tokens.

파라미터

Undisclosed

컨텍스트

라이선스

Proprietary

출시일

2026-04-23

API 가격

입력 가격 (1M 토큰당)

출력 가격 (1M 토큰당)

$30

과금 모드: standard

강점

・Advanced agentic programming capabilities
・Excellent computer operation functions
・Vast one-million-token context

약점

・Lack of transparency from closed-source
・API costs tied to usage
・Limited public licensing model

활용 사례

・Developing autonomous AI agents
・Automating PC operation workflows
・Knowledge work on large documents

심층 분석

Terminal-Bench 2.0

82.7%

SOTA, up from 75.1% (GPT-5.4)

Expert-SWE

73.1%

internal frontier eval

SWE-Bench Pro

58.6%

vs GPT-5.4: 57.7%

Input Price

$5/1M

2x GPT-5.4

Output Price

$30/1M

2x GPT-5.4

Context Window

1M tokens

same as GPT-5.4

강점

・State-of-the-art agentic coding (Terminal-Bench 82.7%)
・More token-efficient than GPT-5.4 — fewer tokens for same tasks
・Matches GPT-5.4 latency despite higher intelligence
・Strong computer use and knowledge work capabilities

약점

・2x price of GPT-5.4 ($5/$30 vs $2.50/$15)
・Trails Claude Opus 4.7 on SWE-bench Pro (58.6% vs 64.3%)
・API availability limited — rolling out gradually

경쟁사 비교

Model	Arena	SWE	GPQA	Price
Claude Opus 4.7	~1500	87.6%	94.2%	$5/$25
GPT-5.5	N/A	N/A	N/A	$5/$30
GPT-5.4	1467	80.8%	94.4%	$2.50/$15

개요

GPT-5.5 is OpenAI's most capable model, released on April 23, 2026. It represents a significant step up in agentic coding, computer use, and knowledge work while matching GPT-5.4's per-token latency — a notable engineering achievement. The model's headline achievement is 82.7% on Terminal-Bench 2.0, a new state-of-the-art for complex command-line workflows. On Expert-SWE, OpenAI's internal frontier eval for long-horizon coding tasks (median 20 hours human completion time), GPT-5.5 outperforms GPT-5.4 significantly. It also uses fewer tokens to complete the same Codex tasks, making it both more capable and more efficient. At $5/$30 per 1M tokens, GPT-5.5 is priced 2x higher than GPT-5.4 but delivers substantially better performance on agentic tasks. For teams building production coding agents, the efficiency gains (fewer tokens, fewer retries) may offset the higher per-token cost.

벤치마크 및 성능

## Benchmark Scores | Benchmark | GPT-5.5 | GPT-5.4 | Claude Opus 4.7 | |-----------|---------|---------|-----------------| | Terminal-Bench 2.0 | 82.7% | 75.1% | 69.4% | | Expert-SWE (Internal) | 73.1% | 68.5% | N/A | | SWE-Bench Pro | 58.6% | 57.7% | 64.3% | | GDPval (wins/ties) | 84.9% | 83.0% | 80.3% | GPT-5.5 delivers state-of-the-art performance on terminal-heavy coding and knowledge work, with significant improvements over GPT-5.4 while using fewer tokens.

상세 비교

## GPT-5.5 vs Claude Opus 4.7 GPT-5.5 leads on Terminal-Bench 2.0 (82.7% vs 69.4%) and knowledge work (GDPval 84.9% vs 80.3%), while Claude Opus 4.7 leads on SWE-bench Pro (64.3% vs 58.6%) and tool orchestration. Both are priced at $5/1M input, but GPT-5.5's output pricing ($30/1M) is higher than Claude's ($25/1M). ## GPT-5.5 vs GPT-5.4 GPT-5.5 offers 7.6 points higher on Terminal-Bench 2.0 and 4.6 points higher on Expert-SWE, while matching per-token latency. The 2x price increase is offset by token efficiency — GPT-5.5 uses fewer tokens to complete the same tasks.

커뮤니티 평가

GPT-5.5 was received with enthusiasm by the coding agent community. The Terminal-Bench 2.2% score was seen as a breakthrough, and early Codex users reported significantly better results on complex multi-file engineering tasks. The pricing at $5/$30 drew some criticism, with developers noting it's 2x GPT-5.4's price. However, OpenAI's argument about token efficiency resonated with production teams — if GPT-5.5 uses 30% fewer tokens, the effective cost increase is closer to 40% rather than 100%. Enterprise adoption has been cautious due to the gradual API rollout, with most teams waiting for full availability before migrating from GPT-5.4.

활용 사례

1. **Production Coding Agents**: Terminal-Bench 82.7% makes it the best model for autonomous coding workflows that involve planning, iteration, and tool coordination. 2. **Complex Knowledge Work**: GDPval 84.9% indicates strong performance on multi-step research, analysis, and document generation tasks. 3. **Computer Use Automation**: Improved OSWorld performance makes it suitable for GUI testing, software automation, and visual workflow tasks. 4. **Scientific Research**: Early testers report strong performance on research partner tasks — critiquing manuscripts, stress-testing arguments, and working with code and PDFs.