PRODUCTS

How OpenAI's GPT-5.6 quietly slipped past Mythos

After Anthropic's Mythos and Fable have dominated the AI headlines in recent weeks, OpenAI has come back to the party. 

On Friday, the company debuted a limited preview of GPT-5.6, the latest iteration of its flagship model. The company unveiled several versions of this model for different levels of tasks, and intends to make them generally available in the coming weeks. 

The company said it previewed its plans and the models' capabilities ahead of launch as part of its "our ongoing engagement with the US government," which is why it has rolled out the model in limited preview to a small group of trusted partners. OpenAI said it will continue testing the model during this period before releasing it to the general public. 

GPT-5.6 has three different classes: Sol, Terra, and Luna. 

• Sol is the "flagship model," representing its strongest iteration yet and proving strong agentic capabilities in coding, biology, and cybersecurity. Sol also includes a new "ultra" mode, which leverages subagents for complex work. As for cost, Sol is $5 per million input tokens and $30 per million output tokens. 

• Terra is a "balanced model" for everyday work, featuring competitive performance to GPT-5.5 while being half the cost. Terra is $2.50 per million input tokens and $15 per million output tokens. 

• Luna, meanwhile, is the cheapest and fastest model in the stack, coming in at just $1 per million input tokens and $6 per million output tokens. 

According to TerminalBench 2.1, which tests workflows requiring planning, iteration, and tool coordination, both Sol and its Ultra mode surpass Claude Mythos 5. Terra surpasses Claude Fable 5 on the same benchmark, while Luna surpasses Opus 4.8. 

Notably, OpenAI strongly emphasized the model family's cyber capabilities and security safeguards in its announcement, noting that all three models are fitted with its "most robust safeguards to date," and configurations that match each model's capabilities. The company said that, as the model becomes more capable, it's designed to hold up against adversarial pressures, while still handling cyber work such as code review, defensive testing, patch development and debugging. 

"Our goal is to make prohibited offensive activity more difficult, uncertain, and detectable without unnecessarily limiting those beneficial uses," OpenAI said in its announcement.

OpenAI's latest addition to the frontier model race represents the latest hop in these companies' trillion-dollar game of leapfrog. And with Anthropic's long-awaited release of Mythos and Fable still caught up in the company's dispute with the government, OpenAI is effectively in a league of its own for now. Though the company is playing it safe, only releasing these models in a limited preview to not stoke the government's ire, the release still gives it a leg up. The fact that GPT-5.6 is outperforming Anthropic's Mythos-class models in several benchmarks definitely doesn't hurt either. However, the bigger question remains: As these companies continue to outdo one another with the latest frontier capabilities, will these upgrades actually make a noticeable difference to users? And, while these models take turns inching past one another, at what point will they become commoditized?

TOGETHER WITH IBM

From weeks of effort to a working application in a weekend

Netherlands-based consultancy Novadoc set out to solve a recurring challenge: configuring FileNet across environments with a more repeatable workflow.

What started as a partially formed framework on a Friday became a working application by Monday morning—work that would normally take at least two weeks.

By analyzing inherited code with limited documentation, the developer generated working documentation and turned the implementation into a configuration management application. The result is a governed workflow that supports extraction, comparison, approval and deployment—reducing manual error and improving reliability when moving business-critical changes across environments.

Dive deeper

RESEARCH

Google makes big bet on AI for science

Can AI create a golden age of scientific discovery? 

In this episode of The Deep View Conversations, we sit down with Yossi Matias, head of Google Research, to explore how AI is transforming the way science gets done.

Rather than replacing scientists, Google is building AI systems designed to amplify human ingenuity. From searching millions of research papers to generating new hypotheses and accelerating experiments, these tools aim to help researchers move from ideas to discoveries faster than ever before.

Yossi explains why he believes we're entering a new era where AI can democratize scientific research, empower the next generation of scientists, and dramatically shorten the path from breakthrough to real-world impact.

Topics covered include:

• How Gemini for Science is changing research workflows

• What AI Co-Scientist, AlphaEvolve, and the Empirical Research Assistant actually do

• Why the scientific method is becoming even more important in the AI era

• How Google is partnering with universities including Stanford and Imperial College

• Why AI could give every researcher a "virtual lab" in their pocket

• What a golden age of scientific discovery might look like

If you're interested in AI, scientific discovery, biotechnology, or the future of innovation, this conversation offers a look at how one of the world's leading AI research organizations sees the next decade unfolding.

📺 Watch on YouTube

🎧 Listen in your favorite podcast player

Subscribe to Deep View Conversations for interviews with the leaders shaping the future of AI, business, and technology: tdv.transitor.fm

TOGETHER WITH CONVEX

Code Faster, Easier, And Less Buggy With Convex

You’re probably already familiar with Convex, the backend building platform that helps your AI agents excel – but their latest updates are taking things up a notch.

Convex recently launched plugins that allow you to connect Claude Code, Codex, GitHub Copilot and more directly to your Convex deployment, allowing them to…

• Read data and run functions entirely on their own

• Optimize your project’s performance via production insights

• Improve safety by limiting access to production

There’s even more updates coming in the near future, but for now, see how you can take your agents’ work to the next level with the latest from Convex right here.

WORKPLACE

AI adoption is outpacing companies' AI strategies

As companies pressure employees to use AI, many are left to fend for themselves.

41% of workers say their employers provide zero guidance on using AI, according to a recent study from career platform ResumeNow. Of the 1,020 U.S. adult workers surveyed, only 21% say their employers offered clear guidelines with specific examples of how to deploy AI in their respective roles.

Formal AI training remains uneven. Only 19% of workers say their employers provided comprehensive AI training with dedicated time and resources. Nearly a third of workers say they receive no AI training at all.

Lack of access to vetted AI tools is widening the gap. More than half of surveyed workers say their employers either provide no AI tools or rely solely on free, publicly available AI models for work-related tasks. As a result, 76% of workers say they've used AI tools they personally found and signed up for to complete work tasks, a trend ResumeNow calls "bring your own AI," or "BYO AI.” 

“When employees are expected to use AI without the right training or access to approved tools, organizations risk creating confusion instead of efficiency,” Keith Spencer, career expert at ResumeNow, told The Deep View. 

The findings come as companies increasingly push employees to use AI in hopes of boosting productivity and cutting costs. As organizations bet big on AI strategies, employees are left scrambling to figure out the best way to incorporate the technology into their work.

“The stakes are high for both employees and leaders,” Spencer said. “For employees, inadequate training can hurt buy-in, damage morale, and increase burnout if they feel overwhelmed by expectations they are not prepared to meet. For employers, limited support can lead to poor ROI, fragmented AI use, and reputational risk if AI-generated or AI-assisted work affects internal decisions, customer-facing materials, hiring resources, or public communications.”

To effectively capture AI's benefits, Spencer suggests employers craft an AI strategy that includes access to approved tools, skills training, and room for employees to experiment.

"Without that structure, AI adoption becomes fragmented and harder to manage," he said.

The findings suggest many companies are treating AI adoption as an employee issue rather than an organizational one. Leaders are telling workers to use AI, but many aren't providing the guidance, training, or approved tools to do it well. Instead, employees are left to their own devices, experimenting with their personal AI accounts and deciding how the technology fits into their workflows. Approaching deployment through trial and error isn’t an AI strategy. It's improvisation. If companies want AI to transform how work gets done and reap the potential gains, they can't expect employees to shoulder the responsibility on their own. Successful AI adoption depends as much on organizational leadership as it does on the technology. And that doesn't even get into the security, privacy, and compliance concerns that have to be part of any successful AI strategy.

LINKS

• Government allows Anthropic's Mythos to redeploy to some US orgs 

• OpenAI poaches Prabhjeet Singh from Uber to scale India presence

• California launches AI job loss tracking tool

• Apple's Vision Pro head Paul Meade is exiting for OpenAI

• General Intuition raises $320 million at $2.3 billion valuation

• China's Z.ai is reportedly matching Anthropic models in cybersecurity

• Luma Connectors: Users can now plug Airtable, Dropbox, Google Drive into Luma. 

• Runway Agent 2.0: A marketing agent that can create marketing briefs, campaign assets and analyze performance data. 

• Google Study Notebooks: The Gemini app now has personalized learning and study capabilities for students. 

• Codex: OpenAI's coding tool is now generally available in the ChatGPT app.

• Zendesk: AI Agent Abuse Prevention Engineer

• Amazon: Applied Scientist, AWS Automated Reasoning

• Anthropic: Software Engineer, Cybersecurity Products

• Paramount: ML Platform Engineer

GAMES

POLL RESULTS

Has there been resistance to AI or agent adoption in your workplace?

Yes (24%)
Somewhat (37%)
No (31%)
Other (8%)