OpenAI released GPT-5.5 on Wednesday, calling it a "new class of intelligence" and the most capable model in its history. The company is charging $200 per month for access through ChatGPT Pro and doubling API prices to $30 per million output tokens. The stock market appeared unimpressed. The AI research community is split.

Here is what actually happened and why it matters.

The Benchmarks: Marginal Gains at Premium Prices

GPT-5.5 scores 93.2% on MMLU-Pro, a graduate-level reasoning benchmark, compared to 91.8% for Anthropic's Claude Mythos Preview and 90.4% for Google's Gemini Ultra 2.5. On Terminal-Bench 2.0, a newer evaluation designed to test real-world agentic coding tasks, GPT-5.5 narrowly edges Claude Mythos by 1.3 percentage points, according to VentureBeat's reporting.

Those are real improvements. They are not the generational leap the marketing suggests. OpenAI's blog post claims GPT-5.5 is "approximately twice as capable" as GPT-5.4, which launched just weeks ago. But the benchmark spreads between top models have compressed to single-digit percentages. The frontier is crowded, and the price to sit at the front just doubled.

What GPT-5.5 Can Actually Do

OpenAI is positioning this model as the backbone of its "super app" ambitions. According to TechCrunch, GPT-5.5 can now handle multi-step workflows that previously required chaining multiple API calls: booking travel while cross-referencing calendar availability, debugging code while simultaneously writing documentation, analyzing financial reports while generating presentation slides.

The model supports a 1-million-token context window (up from 256K in GPT-5.4) and introduces what OpenAI calls "persistent memory threads" that maintain context across sessions without explicit user prompting. Native image generation is built directly into the model rather than routing to DALL-E, which OpenAI says reduces latency by 60%.

For developers, OpenAI is rolling out function-calling improvements and a new "orchestration mode" that lets GPT-5.5 manage sub-agents autonomously. Sam Altman wrote in a blog post that the company envisions GPT-5.5 as "the last model most people will need to interact with directly," with future models running behind the scenes as specialized workers.

The Pricing Problem

This is where it gets uncomfortable for OpenAI. The Decoder reported that API pricing has doubled across the board: $15 per million input tokens and $30 per million output tokens. ChatGPT Plus ($20/month) users get limited access. Full GPT-5.5 capabilities require ChatGPT Pro at $200/month.

Anthropic's Claude Mythos, which scores within 1-2 percentage points on most benchmarks, charges roughly half that for API access. Google's Gemini models are cheaper still. For enterprise customers running millions of API calls per day, GPT-5.5's marginal benchmark advantage may not justify a 2x cost increase.

OpenAI is betting that capability matters more than cost at the frontier. That bet has worked before. It may not work in a market where Anthropic and Google are shipping competitive models at lower price points every few weeks.

The Safety Question Nobody Is Asking

GPT-5.5 launched just weeks after GPT-5.4. Before that, GPT-5.3 arrived in March. The cadence is accelerating. OpenAI's safety team, which has lost several senior members over the past year, is now evaluating models on compressed timelines.

OpenAI says GPT-5.5 passed all internal safety evaluations with "the highest marks of any model we've released." The company did not publish the full evaluation framework or invite external red-teaming before launch. Anthropic, by contrast, published its Mythos safety card two weeks before general availability.

What This Means

GPT-5.5 is a real product with real improvements. The agentic capabilities are genuinely new. The context window expansion matters. The persistent memory threads could change how people use AI daily.

But "new class of intelligence" is doing a lot of heavy lifting for what the benchmarks show is a strong incremental improvement at a premium price point. The Polymarket prediction crowd had this pegged at 86% for an April 23 drop, and they were right. The question is whether the market for $200/month AI subscriptions is as predictable.

OpenAI is the most valuable private company in the world. GPT-5.5 is its strongest model. But the gap between "strongest" and "strong enough" is shrinking every month. And that gap is where Anthropic, Google, and the open-source community live.