BIG TECH

Gemini Carbon tests Google’s ability to keep pace

Just weeks after releasing its latest frontier model, Google has stepped up to the plate with another one designed specifically for security and heavy agentic workloads. 

On Wednesday, the tech giant released Gemini Carbon claiming state-of-the-art performance on cybersecurity and long horizon capabilities. The company said Carbon offers substantial gains on coding, agentic tasks and key capabilities for cyber defense, as well as shows decreases in misalignment and "fabrication rates," or the rate that it creates false responses, in internal testing on living code environments. 

The model builds on Argon, the company's frontier model unveiled in late September. The company said that it worked with the US government's voluntary framework for pre-release model testing, and will make the model available to a "subset of cyber defenders" starting this week, gradually extending access to partners through its Fairwind program over time before making it generally available. 

 "Carbon illustrates the pace at which we’re pushing the frontier forward," Google said in its announcement. "Our researchers and models are working hand in hand to advance capabilities faster, while strengthening our safeguards in tandem." 

Google touted a number of safety and security advancements with Carbon, noting that it reduced the model's cheating and shortcut attempts, cut unsupported claims by two times compared to its predecessor Argon, resisted "every evaluated attack attempt" for prompt injection, and cut down overreach to keep it from breaking free of constraints. 

As far as benchmarks, Carbon sits at the top of the Artificial Analysis Intelligence Index, which itself was updated on Wednesday to reflect updates to existing benchmarks and the addition of two new evaluations for coding and science tasks. 

• Carbon just outranks Claude Opus 5.5 and GPT-6 Astra in the index, scoring a 59 to their respective 58 and 57. Additionally, it sweeps these competitors on benchmarks for software engineering, agentic tasks, exploit finding, knowledge work, and more. 

• It only falls behind competitors on two benchmarks: OSWorld, a benchmark testing multimodal AI agents on open-ended computer tasks, and FrontierSWE, which evaluates ultra-long-horizon software engineering. 

Carbon will launch at the same introductory price as Argon, sitting at $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price. That's on par with OpenAI's GPT-6.1 Sol and Athropic's Claude Sonnet 5.5, which each cost $2 per million input, $10 per million output, and 10 cents per million cached reads.

As we've said before, Google still has all the ingredients to win in AI, with a trusted brand legacy, large existing enterprise market share, and practically boundless cash and compute resources. But Google hasn't been able to keep up with OpenAI and Anthropic on speed. Throughout 2026, the two upstart labs have raced to churn out new and improved models at furious pace. Google, meanwhile, has struggled to release a high-end model all year, instead releasing niche and efficiency-focused models that target lower end enterprise tasks. However, Carbon could be a sign that the company is seeking to pick up speed, even as the industry faces broader calls to slow down the pace of the frontier.  

TOGETHER WITH NOOKS

Sales is harder than code. Nooks Labs is working on it.

Coding agents get feedback in seconds: the tests pass or they fail. Nooks learned that sales agents can wait weeks to find out whether a move worked. Much of what matters, like who's really making the decision, never makes it into the CRM.

Nooks' agents scored 97% on the company's own eval set, but reps still double-checked every output. So Nooks Labs, the company's applied AI team, built a harder benchmark from real deal data. The team then fixed what it exposed with calibrated confidence scores, a way to test a strategy against past deals before it runs, and 500+ examples of bad AI writing used as guardrails.

Today, 96% of agent-written emails go out without edits. The open question now: what does it take to go from 2x productivity gains to 10x?

Discover the research →

PRODUCTS

GitHub unveils fix for AI's token economics problem

In a single move, GitHub may have solved one of AI's most pressing issues and put itself back into the coding agent race.

On Wednesday at its GitHub Universe event in San Francisco, the company unveiled GitHub Copilot Fusion, a multi-model orchestration feature that can save time and tokens while potentially achieving results that match using high-end frontier models alone. 

That will be music to the ears of many CFOs and business executives who are sweating over ballooning AI costs. As we reported last week from Gartner Symposium, 81% of CIOs report that they are increasing AI spend while only 11% of CFOs report that they are seeing clear returns from AI.

Copilot Fusion potentially reduces costs by automatically routing tasks from AI builders to less expensive models. The way it works is that you can use the GitHub Copilot app as your coding harness and select Copilot Fusion as your model. Then you select Standard, High, or Max for the amount of effort you think your task needs and the software routes to different agents and subagents to complete the task.

There's a lot of nuance and complexity to the way it works, but the bottom line is that it starts with simpler models and escalates up as needed. The result, based on the Artificial Analysis Coding Agent Index, is Copilot Fusion Max is about as powerful as Claude Code with Opus 5.5 Max while using half as many tokens and using 40% less time to complete the average task.

GitHub COO Kyle Daigle told The Deep View that GitHub’s internal teams and early enterprise adopters are seeing roughly 50% to 60% lower costs on complex work with Copilot Fusion, and some are seeing up to 77% lower costs.

"We're really excited about it because it resolves the question of 'Do I really need this frontier intelligence?' And the answer is 'No, not all the time,'" Daigle said. 

In addition to Copilot Fusion, GitHub also made several other key announcements at the event, including:

• Projects: Offers a persistent workspace across goals, context, sessions, and results to coordinate work and track progress.

• AI Scans: Helps find vulnerabilities in code, determines whether code calls vulnerable functions, and blocks known malware and critical vulnerabilities.

• App switching: You can now move between your phone and desktop apps to handle many of the functions of Copilot.

Coding agents have been a red-hot AI topic for 2026. But while GitHub Copilot is one of the incumbents in the coding agent space, it has largely been overshadowed by Claude Code, OpenAI's Codex, and SpaceXSI's Cursor throughout the year. With Copilot Fusion, the company is playing to its strengths and giving itself an excellent chance to get back in the game. Claude Code and Codex are never going to be likely to route to less expensive models, open models, or small language models because those companies are incented to maximize your model spend. GitHub Copilot, on the other hand, is incented to align with developer preferences. And Daigle told us that GitHub is also making it so that you can choose your own local models as an option in Copilot Fusion as well—increasing the speed and decreasing the costs even further. That gives GitHub an opportunity to help solve token economics, which is one of the most entrenched problems in AI right now.

TOGETHER WITH WIZ

A Practical Framework for AI Threat Readiness

As attackers use AI to identify vulnerabilities and accelerate exploitation, security teams need operational strategies that keep pace.

Wiz’s AI Threat Readiness Playbook provides a four-pillar framework security leaders can use to strengthen programs and prepare for AI-enabled threats.

Learn how to:

• Reduce critical attack surface exposure

• Improve zero-day response and remediation

• Strengthen application security with AI-assisted analysis

• Modernize detection and response using AI-driven workflows

Get the Playbook

GOVERNANCE

How AI is forcing cybersecurity to evolve

The AI agents helping employees work faster also open up new risks for a company's most sensitive data. So how do businesses keep up with the opportunities and the risks?

In this episode of The Deep View Conversations, we sit down with Christian Rodriguez, CrowdStrike's field CTO for the Americas. We talked to Rodriguez at CrowdStrike's Fal.Con 2026 event to explore what one of the world's largest cybersecurity companies is learning about how AI is rapidly changing the cybersecurity landscape.

Rodriguez explains why defenders need to respond at machine speed, what the OpenAI and Hugging Face incident reveals about goal-seeking agents, and why even useful AI tools need visibility, boundaries, and accountability. Rodriguez argues that data governance remains a central challenge: companies need to understand what their agents can access, where information moves and who is responsible for each action.

Topics covered:

• Helping healthcare, government and other understaffed organizations respond to faster attacks
• Whether automation and tool consolidation can strengthen security within existing budgets
• SafeMind, red and blue cybersecurity models, and the role of the AI harness
• Bring-your-own-model options and where CrowdStrike's models run
• Moving from reactive incident response toward proactive security with human oversight
• How Rodriguez uses Slackbot to organize his day and follow through on commitments 

If you're leading an organization, putting agents to work, or trying to understand how AI changes the security and safety landscape, this conversation connects the new capabilities with the controls and judgment needed to manage them.

📺 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: https://tdv.transistor.fm/

LINKS

• Physical AI startup Foundry raises $100 million for autonomous factories

• OpenAI launches election "war room" ahead of midterms 

• Dropbox to cut 9% of staff, CTO Ali Dasdan to step down

• AI security startup Upwind raises $400 million, $3.8 billion valuation

• Lawsuit alleges ChatGPT gave incorrect medical advice, leading to death

• NY Gov. Hochul bans AI chatbots from using "unsafe" features for kids

• Claude Enterprise: Anthropic's enterprise tool now has smart reports, in-app surveys and pooled budgets. 

• Artificial Analysis Intelligence Index v5: The benchmarking index now features two new evaluations for coding and agentic science. 

• LFM3: Liquid AI's new generation of foundation models across text, audio, vision, and omni modalities. 

• MAI-Image-2.6: Microsoft's image model and flash version are now generally available on Microsoft foundry. 

• Digital Waffle: AI Engineer

• Fanatics: Applied AI Engineer

• DeepRec.ai: AI Engineer

• OpenAI: Applied AI Architect

GAMES

POLL RESULTS

Do you think your organization has good visibility into the actions of it's AI agents?

Yes (20%)
No (46%)
I’m not sure (29%)
Other (5%)