Hi tech & AI enthusiasts, this week's highlights:
xAI Released Grok 4.6 and Cut the Price in Half
DeepSeek Launched a New Model and an Official Open Harness
OpenAI Previewed a 14x Speed Boost for Top Models
The Lab: How to Set Up Hermes with OpenRouter and DeepSeek
AI Image of the Week: Breaking free from AI vendor lock-in
Let's get into it.
▲ AI MODELS
xAI launched Grok 4.6, its new flagship model. It lands right alongside the industry leaders from Anthropic and OpenAI while costing less than half of what they charge. A year ago Grok lagged behind. This week it caught the frontier.
What you need to know
A punchline no longer: Grok 4.6 sits just behind Claude's top model and passes OpenAI on major reasoning tests. On real-world legal and professional work it edges ahead both.
Less than half the bill: xAI is charging 60% less than rival frontier models for the same work.
Grok 4.7 is already on the way: The next upgrade is stated to ship within 3-4 weeks, targeting the top spot across all performance categories.
Key Idea
▲ Why it matters: When a top AI model costs 60% less, the pricing floor moves for everyone. If OpenAI and Anthropic want to keep charging premium rates, they now have to justify why they are worth double. That competition on price and capability is what lets businesses access better models for lower costs.
▲ OPEN MODELS
DeepSeek released a new model update for DeepSeek V4 Pro alongside an official harness (agent software) designed to make deploying and running open models easier for teams.
What you need to know
New model release is cheap: DeepSeek's newest release is smarter and has faster response time. It’s the go to low cost AI option available.
An official harness to run it: Alongside the model, DeepSeek shipped its own dedicated harness so businesses can connect and manage models easier. Think OpenAI codex/Claude but you can plug in any AI model.
Key Idea
▲ Why it matters: AI performance keeps getting better while open models like DeepSeek are keeping prices down. The open harness makes cheaper open-weight AI models a viable option for businesses to use for tasks & automations. If you’ve building on top of OpenAI or Anthropic models it’s worth testing these models out to save yourself and business money.
▲ AI INFRASTRUCTURE
OpenAI previewed Ultrafast, a new tier powered by Cerebras chips that speeds up answers from its flagship model by up to 14 times.
What you need to know
"Like genuinely cheating at my job": That is how one OpenAI staffer described the pace. Another said complex security investigations that used to take hours now finish in ten minutes.
11 hours instead of 78: On a massive 2,500-question research benchmark, the Ultrafast setup finished in under half a day. Rival frontier setups took more than three days to complete the exact same test.
Invite-only for now: Access is in private preview with no pricing listed yet. OpenAI says it will open up as more specialised hardware comes online.
Key Idea
▲ Why it matters: 14x speed is not about getting a chatbot reply a few seconds faster. It changes what autonomous agents can do. When an AI can test 20x approaches in the time it used to take to test one, complex research and troubleshooting stop being overnight batch jobs and start happening live. I wouldn’t be surprised if we will start seeing more AI models targeting faster speeds over the months ahead.
▲ EVEN MORE NEWS
Other big ideas from the past week.
▲ OpenAI released Computer History: Turn recent computer activity into memories and a timeline that ChatGPT and Codex can use.
▲ xAI Put Grok Bot into Beta: xAI launched Grok Bot, positioning it as an AI teammate that handles multi-step operational tasks in the background rather than just answering prompts.
▲ Z.ai Shipped GLM-5.3: Chinese lab Z.ai released GLM-5.3, upgrading its open model with sharper coding and multi-language comprehension.
The Lab
▲ AI EDUCATION
Relying on one AI provider is risky. If prices rise, limits drop, or a server goes down, your work stops. This is why using an open harness like Hermes with OpenRouter lets you switch between DeepSeek, Grok, GPT, and Claude in two clicks.
What you'll learn:
Why separating your chat interface from your model provider saves money and headaches
How to set up one OpenRouter key to access hundreds of models
How to configure Hermes to run DeepSeek's newest models and official harness tools
Which model to use for routine tasks vs deep reasoning
🔬 Interested in learning about AI for business? Join The Lab waitlist.
▲ AI IMAGE SHOWCASE
Create this type of image with Nano Banana 2
A seamless 50/50 split-view architectural photograph of a modern luxury building. The left half shows the active construction phase with precision-engineered raw steel beams, structural timber framing, and smooth polished concrete slab under bright crisp morning sunlight. The right half seamlessly transitions into the completed space featuring warm Scandinavian oak cabinetry, minimalist floor-to-ceiling glass walls, recessed warm architectural lighting, and designer furniture. Straight-on eye-level perspective, crisp focus, photorealistic architectural photography, shot on a Hasselblad 50mm lens at f/5.6.Image Usage Suggestion: Case study slide decks, construction and architecture portfolio headers, transformation carousels, client proposal documents, LinkedIn project breakdown posts, or before/after website hero banners.
💌 Reply to this email with your AI image generation or suggestion of what you'd like to see and I'll feature it in a future newsletter.
▲ SHARE YOUR THOUGHTS
Help shape the content you see here by giving feedback
Have specific feedback or want to get in touch? Reply to this email and I’ll get back to you.
Know someone who’d love this newsletter? Forward it to a friend and have them sign up here.
Thanks for reading, until next time.
Stay curious,
Matt Lok, Editor
Brought to you by Metalabs. Digital marketing consulting specialising in AI. Carousels made with SlideDrift.