OpenAI releases cyber-capable models as Google cuts AI pricing
OpenAI is making high-level cybersecurity tools available to more users while Google and others slash costs for financial and legal tasks.

The short version
- OpenAI released GPT-6.2 Astra and Sol, which meet the company's 'Critical' cybersecurity threshold.
- Google's Gemini 4 Neon costs $3 per million input tokens, less than half the price of OpenAI's top models.
- Tencent's Hy4 Flash is priced at just 7 cents per million input tokens.
- OpenAI introduced a 'Control Room' monitoring system to oversee the use of these new cyber-capable models.
Why it matters
The release of cyber-capable models means that tools for finding and exploiting software flaws are becoming more accessible to the general public. At the same time, falling costs allow businesses to automate complex tasks like contract review and financial modeling at a much lower price point.
OpenAI's Cyber Capabilities
OpenAI officially released two new models, GPT-6.2 Astra and GPT-6.2 Sol, on November 17. Both of these models meet the company's 'Critical' cybersecurity threshold, which is a specific classification for AI that can perform high-level security tasks. This classification previously led the company to restrict access to similar models for only approved customers.
Instead of keeping these capabilities behind a closed door, OpenAI is opening them up to a wider audience. To manage the risks associated with this move, the company introduced a monitoring system called Control Room. This system is intended to oversee how users interact with models that can identify and exploit software vulnerabilities.
The release effectively turns sophisticated cyber-offensive capabilities into a commodity. By making these tools available, OpenAI is shifting the landscape of how security teams must defend against potential threats. Organizations now have to assume that attackers may have access to the same high-powered tools that security professionals use for defense.
Google's Price Cuts
While OpenAI expanded its capabilities, Google focused on making its AI more affordable for professional work. The company launched Gemini 4 Neon, which is priced at $3 per million input tokens. This price point is less than half of the $8 charged by OpenAI and Anthropic for their top-tier models.
Google designed Gemini 4 Neon specifically for accounting, spreadsheet modeling, and tax work. It ranked first on several benchmarks for legal research and tax agent tasks. For businesses, this means a firm can perform contract reviews using a top-ranked model for about one-third of the previous spend.
The company also released Gemini 4 Hydrogen, which is priced at half the cost of the Neon model. This model is designed for speed, completing tasks 2.5 times faster than previous versions. It is also integrated into Google Sheets for business subscribers, placing AI directly into the tools that finance teams use every day.
The Race to the Bottom
Other companies are joining the push to lower the cost of AI tokens, which are the small chunks of text used to measure AI usage. Tencent launched Hy4 Flash, which costs less than one-hundredth of the price of OpenAI's Astra model. In internal tests, Tencent reported that user satisfaction with this low-cost model approached the levels of much more expensive competitors.
Alibaba also entered the fray by releasing Qwen4-Max. The company reported a roughly 60% reduction in average cost per task compared to its previous model. They also noted that a manufacturing customer used the technology to cut the time required for programming machine parts by 95%.
These price cuts are creating a situation where cost is no longer a primary barrier for many industries. Financial institutions can now deploy AI across a wider range of workflows, such as financial modeling and investment memos, without the bill deciding whether or not to use the technology.
Decision Models and Speed
A new category of AI called decision models is also emerging to handle specific business workflows. Unlike standard models that write text, decision models select from predefined options and provide a confidence score. These are used for tasks like routing service requests, screening transactions, or checking if a submission meets company policy.
Google launched Gemini Decision 1, which can evaluate up to 1,000 questions in less than a quarter of a second. It operates at roughly one-eightieth the price of the Astra model. This low cost makes it feasible for companies to screen every single transaction or document with AI rather than just sampling them.
Microsoft followed suit by releasing Microsoft-Decision-1.1. This model was built on Gemma 4, which is an open model from Google. Microsoft stated that the model can decline to make a decision if its confidence is low, which is a critical feature for compliance workflows where a wrong decision is more costly than no decision at all.
What to watch
- How the OpenAI Control Room system handles requests to exploit specific software vulnerabilities.
- Whether the low cost of decision models leads to a standard practice of screening every transaction with AI.
Try it
Check if your company's current AI usage is being metered by tokens and see if switching to a lower-cost model like Gemini 4 Neon could reduce costs for routine tasks like contract review.
Sources and links
- Officialopenai.com/index/introducing-gpt-6-2-astra-sol/ openai.com
- Officialblog.google/intl/en-in/products/introducing-gemini-4-hydrogen/ blog.google
- Officialblog.google/innovation-and-ai/technology/developers-tools/gemini-decis blog.google
- Reportingpymnts.com: the article we sourced
First reported by pymnts.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.