OpenAI Releases GPT-5.5: Faster, Smarter—And Pricier

OpenAI launched GPT-5.5 connected Thursday, pitching it fundamentally arsenic a exemplary targeted astatine agentic machine use. It writes and debugs code, browses the web, fills retired spreadsheets, and keeps moving done multi-step tasks without needing a quality to babysit each move.

The merchandise is already rolling retired contiguous to Plus, Pro, Business, and Enterprise subscribers crossed ChatGPT and Codex, OpenAI said.

“We’re releasing GPT‑5.5, our smartest and astir intuitive-to-use exemplary yet, and the adjacent measurement toward a caller mode of getting enactment done connected a computer,” OpenAI said successful an announcement. “The gains are particularly beardown successful agentic coding, machine use, cognition work, and aboriginal technological research—areas wherever advancement depends connected reasoning crossed discourse and taking enactment implicit time.

The large header from OpenAI: GPT-5.5 is measurably smarter than its predecessor, GPT-5.4—and it's not slower. Matching GPT-5.4's per-token latency successful real-world serving portion hitting higher scores crossed benchmarks is the benignant of ratio betterment that usually doesn't happen. Bigger models thin to beryllium slower erstwhile moving nether the aforesaid hardware.

On Terminal-Bench 2.0, which tests however good a exemplary handles analyzable command-line workflows that necessitate readying and iterative instrumentality use, GPT-5.5 scores 82.7%. Claude Opus 4.7 lands astatine 69.4%, portion Gemini 3.1 Pro sits astatine 68.5%. That's not a marginal lead.

On GDPval, a benchmark investigating cognition enactment crossed 44 existent occupations—from concern to ineligible probe to merchandise management—GPT-5.5 matches oregon beats manufacture professionals successful 84.9% of comparisons.

Image: OpenAI

It’s besides a beauteous bully coder, arsenic expected. On Expert-SWE, an interior benchmark for long-horizon coding tasks with a median estimated quality completion clip of 20 hours, GPT-5.5 outperforms GPT-5.4. On SWE-Bench Pro, which grades real-world GitHub contented resolution, it reaches 58.6%. Claude Opus 4.7 scores higher astatine 64.3%, but OpenAI claims it whitethorn beryllium due to the fact that “Anthropic reported⁠ signs of memorization connected a subset of problems”

This motorboat lands successful a marketplace that is moving rapidly since the roar of agentic AI. GPT-5.4 arrived conscionable 2 days aft GPT-5.3, portion Xiaomi went from MiMo-V2-Pro to MiMo 2.5 Pro—with afloat multimodal capabilities—in astir 5 weeks. The spread betwixt GPT-5.4 and GPT-5.5 was astir 7 weeks. That's the tempo now.

But volition this exemplary marque a quality for mundane users who are not ever coding the adjacent large thing? If you're connected a escaped tier, no: GPT-5.5 isn't coming to escaped users. If you're paying for Plus astatine $20/month, it rolls retired today. We tried investigating it nether our Pro account, but the exemplary was not instantly available.

The bigger woody is astir apt what GPT-5.5 does wrong Codex—OpenAI's agentic coding environment—where it is proven to beryllium much powerful. “It genuinely feels similar I’m moving with a higher intelligence, and there’s astir a consciousness of respect,” Pietro Schirano, CEO of MagicPath, said successful a punctuation shared by OpenAI.

GPT-5.5 Pro, designed for harder, higher-accuracy work, is rolling retired separately to Pro, Business, and Enterprise users successful ChatGPT. On BrowseComp, which tests a model's quality to way down hard-to-find accusation crossed the web, GPT-5.5 Pro scores 90.1%, up of Gemini 3.1 Pro astatine 85.9%.

The exemplary is besides the astir intelligent connected mean based connected the Artificial Analysis Index. GPT 5.5 reports a much businesslike and utile usage of tokens, producing amended results successful general.

Image: OpenAI

The pricing, however, could daze immoderate users. The API volition complaint $5 per cardinal input tokens and $30 per cardinal output tokens erstwhile it launches, which OpenAI says is coming "very soon." GPT-5.5 Pro successful the API volition outgo $30 per cardinal input tokens and $180 per cardinal output tokens.

Those figures are higher than GPT-5.4—$2.50 per cardinal tokens of input and $15.00 per cardinal tokens of output—while pricing for GPT-5.5 Pro remains the aforesaid arsenic GPT-5.4 Pro.

That said, OpenAI CEO Sam Altman argued connected X that token ratio gains offset the cost—GPT-5.5 completes the aforesaid Codex tasks with less tokens, which means cheaper runs adjacent astatine a higher per-token rate.

Just for comparison, Xiaomi MiMo v2.5 Pro charges $1 and $3 per cardinal tokens of input and output, Minimax M2.7 costs $0.30 and $1.20 respectively, and Kimi K2.5 requires $0.44 and $2.00 per cardinal tokens.