SunBrief#93: OpenAI Makes GPT 5.6 Sol 14× Faster

Welcome to the SunBrief

Today in SunBrief 🌞

• You Can’t Scale If You’re Still Doing Everything

• OpenAI Previews GPT 5.6 Sol Ultrafast at Up to 14 Times the Speed

• Stock Updates

• Google Launches Gemini 3.7 Flash With Stronger Coding at Half the Price

• SpaceXAI Launches Grok 4.6 for Longer Agent Work and Visual Projects

• AI Highlights of the Week

• Too Important to Miss

You Can’t Scale If You’re Still Doing Everything

As your business grows, your role should evolve with it.
But for many leaders, it doesn’t.
They stay buried in tasks, decisions, and responsibilities they’ve already outgrown.
BELAY created the free resource From Operator to Owner to show leaders how to step out of the day-to-day without losing control and what it takes to operate at the next level.
At BELAY, we match you with U.S.-based Assistants who take on the operational work that’s keeping you stuck.

OpenAI Previews GPT 5.6 Sol Ultrafast at Up to 14 Times the Speed

Cerebras helps deliver up to 750 tokens per second for work where every second matters

OpenAI is testing a new Ultrafast tier for GPT 5.6 Sol that runs the same frontier model at up to 14 times Standard speed. Powered by Cerebras, it can generate as many as 750 tokens per second, bringing Sol into workflows where a slow answer can be almost as costly as a wrong one.

Key Points:

• Sol Without the Wait: Ultrafast runs GPT 5.6 Sol up to 14 times faster without reducing the model’s intelligence.

• Cerebras Powers the Speed: The service can generate up to 750 output tokens per second using Cerebras inference hardware.

• Built for Live Work: OpenAI sees the biggest gains in outages, fraud detection, customer support, commerce, and fast research cycles.

• OpenAI Is Testing It Too: Internal teams use Ultrafast to investigate incidents and turn overnight research loops into same-day work.

• Limited Preview: Access is currently restricted to selected API customers and will expand as more computing capacity becomes available.

Why It Matters:
Ultrafast makes frontier AI practical in moments where waiting a minute can cost money, customers, or uptime. If OpenAI can scale the service, businesses may no longer need to choose between a smart model and a fast one for live support, coding, finance, and incident response.

Stock Updates

Google Launches Gemini 3.7 Flash With Stronger Coding at Half the Price

The new workhorse model improves software engineering, web development, and business agents

Google has released Gemini 3.7 Flash just three weeks after 3.6 Flash, with a sharper focus on coding, web development, and business automation. The company says it gets more right on the first try, follows instructions more closely, and needs fewer retries to finish complex work.

Key Points:

• Coding Takes a Leap: The model scored 65.3% on DeepSWE, up from 49%, while improving debugging and first pass code accuracy.

• Better Websites, Fewer Prompts: It follows screenshots and design systems more closely, producing more complete interfaces with less back and forth.

• Serious Work Gets Easier: Google says it is better at reading dense documents and completing workflows across finance, law, and biosciences.

• Half the Price: Through the end of 2026, it costs $0.75 per million input tokens and $3.75 per million output tokens.

• Spark Gets Smarter: Gemini Spark now uses 3.7 Flash to manage files, draft emails, and update Workspace documents with better accuracy.

• Available Across Google: Developers can access it through the Gemini API, AI Studio, Android Studio, and Antigravity, with enterprise access through Gemini Enterprise.

Why It Matters:
Gemini 3.7 Flash gives developers a stronger model without forcing them into flagship pricing, making capable coding and business agents cheaper to run at scale. Its rapid arrival after 3.6 Flash also shows how quickly Google is turning developer feedback into updates across Gemini and Workspace.

SpaceXAI Launches Grok 4.6 for Longer Agent Work and Visual Projects

The new model stays with complex tasks longer and turns broad ideas into polished working products

SpaceXAI has released Grok 4.6, an update built for the kind of work that often causes coding agents to lose the thread. It is designed to research, plan, build, and refine projects across longer sessions, with stronger results in software engineering, business work, and interactive design.

Key Points:

• Built for Longer Work: Grok 4.6 can stay with research, coding, and analysis tasks across many steps without stopping too early.

• Better First Versions: It can turn a broad product idea into a working app, then improve the design and interactions through feedback.

• More Self-Checking: The model tests and verifies its own work more often before moving on, reducing the need for constant supervision.

• Back at the Frontier: Artificial Analysis places it alongside GPT 5.6 Sol overall, though Anthropic’s top model still holds a slight lead.

• Aggressive Pricing: API access starts at $2 per million input tokens and $6 per million output tokens, far below many frontier rivals.

• Available Now: Grok 4.6 is live in Cursor, Grok Build, the SpaceXAI API, and several major model platforms.

Why It Matters:
Grok 4.6 gives developers a lower-cost frontier model for projects that need hours of reasoning, coding, and revision without constant supervision. If its real-world performance matches the launch claims, SpaceXAI could become a stronger rival to OpenAI and Anthropic in coding tools and business automation.

AI Highlights of the Week

• Alibaba Releases Qwen3.8-Max AI Model

  Alibaba launched Qwen3.8-Max, a 2.4T-parameter MoE model with 95B active parameters and a 1M-token context window.

  The model targets coding, research, and long-running agent tasks, and its full weights are now publicly available.

• Google Lets Gemini Users Remove Visible AI Watermarks

  Google is rolling out a new setting that lets users turn off visible watermarks on AI-generated images, videos, and music.

  The content will still contain invisible SynthID watermarks and C2PA metadata, so it can still be identified as AI-generated.

• DeepSeek Launches V4 Pro at Up to 14× the Price

  DeepSeek launched V4 Pro, its new flagship AI model with stronger reasoning and agent capabilities than V4 Flash.

  But that extra performance comes at a steep cost, with V4 Pro priced at up to 14× more than V4 Flash.

• Google Gives Pixel 11 an AI Camera That Shoots for You

  Google introduced Magic Capture, which uses Gemini and on-device AI to automatically capture perfectly timed photos on the Pixel 11.

  The AI analyzes around 400 frames, picks the best moment, and can automatically crop and unblur the final shot.

Too Important to Miss

China’s Z.ai claims top open coding AI

Apple trained a China AI with Alibaba

ChatGPT remembers your screen history

Google DeepMind Brings Sign Language AI to Pixel 11

Google makes Quick Share even more like Apple's AirDrop

Last Week’s Poll Result

• Does Ask Maps turn Google Maps into more than a navigation app?

  Yes, it is becoming an everyday AI agent → 42.86%

  Somewhat, but navigation is still core → 28.57%

  No, these are just extra features → 28.57%

• Does unlimited GPT-5.6 Luna access make the free version of ChatGPT significantly better?

  Yes, definitely → 37.50%

  Somewhat → 50.00%

  Not really → 12.50%

• What does this research signal about AI and biology?

  AI is becoming a true biological designer → 33.33%

  Drug discovery could accelerate dramatically → 11.11%

  Biosecurity is becoming more urgent → 55.56%

Feedback

Know someone who may be interested?