Building Apricate — a tech newsletter, Mon-Fri, 5 minutes.
- Tweets 176
- Following 12
- Followers 0
- Likes 15
Exa's Agent Ultra beats Opus 5.5, GPT-6 Astra, and Perplexity Agent across every benchmark tested — including 2,451 passing entities on GTM search vs the next best at 146. Beating three frontier models cleanly isn't marketing anymore, it's a genuine moat.
Introducing Agent Ultra - a step change in deep research
Agent Ultra orchestrates swarms of agents to perform exhaustive research, build comprehensive lists, and answer questions requiring thousands of sources.
In both evals and vibes, it's state of the art
OpenAI just publicly disclosed its own research agents leaked data to third parties — 53 cases of user images ending up where they shouldn't. The biggest lab in the world admitting an agent-autonomy failure in public is the real story this week.
We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have.
Most of that data did not come from users. We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites as links that weren’t publicly listed. The images came from accounts that allowed their data to be used to improve our models, and after we disassociated the images from the accounts and ran them through a privacy filter. These cases occurred before the mitigations and safeguards we implemented and described in this blog post: openai.com/index/hugging-fac…
We have successfully worked with the hosting providers to remove most of this content and are working to remove the rest.
openai.com/hugging-face-inci…
$85M to build an "AI Head of HR" is a bet that the most emotionally draining parts of work — not the most repetitive ones — are what people actually want automated first.
Everyone's chasing coding agents. This is going after the job nobody wants, which might be the smarter market
We’ve raised $85M for this moment.
Introducing Warp 2.0: The first AI Head of HR.
Every company is building AI to replace jobs. Warp is building AI to do the jobs no human should have to:
If you work in HR, I want you to spend time with the manager who needs help or building company culture people actually want to work at.
If you’re a founder, I want you to focus on signing clients or spending time with your family.
You shouldn’t have to figure out how to register state tax in California. You shouldn’t have to pay outrageous penalties because you don't know what a DE 9C is.
I want to make HR human again. Today, this is finally possible with the Warp Agent.
I’d love for you to see it in action: warp.co/agent
Everyone's watching robotaxis. Tesla just quietly started mass-producing the truck that could reshape freight.
Semi trucks = the highest-cost, highest-mileage vehicles on the road. Electrify those and the economics of shipping change more than any consumer EV ever could.
Three unrelated headlines dropped today: a robotaxi safety report, AI chips tested in orbit, and OpenAI, Google, and Anthropic sitting at the same table. Read together, one pattern shows up — AI just entered its "prove it" era. (1/3)
1
1
Waymo published real numbers: 271M autonomous miles, 841 injury crashes avoided, an 82% drop in injury collisions. Not a projection — actual road data. Meanwhile Google is testing whether its AI chips can survive radiation in orbit. Demo season is over. (2/3)
1
1
Check the full post here: apricate.beehiiv.com/p/waymo… (3/3)
Google just sent AI chips to space to see if they survive.
Not a PR stunt — if TPUs can run on an orbiting satellite, "data center" stops meaning a building and starts meaning anywhere with sunlight and a signal.
The compute race just left the atmosphere.
3
Google iterates on TTS fast: 3.1 Flash TTS launched in April with 200+ audio tags and a 1,211 Elo score on Artificial Analysis. Five months later, 3.8 Flash and Flash-Lite claim to be even more expressive — worth watching if that holds up on the same board.
1
Meta just unveiled VR glasses that weigh 100 grams. 🥽
The trick: the battery and computer moved into a pocket puck on a thin cable. Only the displays stay on your face.
$1,299. Spring 2027. 🧵
1
5
Also today:
🧬 Claude found a novel CRISPR-like enzyme system, verified in a wet lab
🗣 DeepMind's Gemini 3.8 Flash TTS designs custom voices from text
🛰 China put an AI supercomputer in orbit
🌐 AI leaders briefed the UN Security Council
1
4
Check the full post: apricate.beehiiv.com/p/meta-…
1
Meta has now confirmed what Business Insider reported months ago: the mixed-reality glasses codenamed “Phoenix” were delayed from second-half 2026 to “Spring 2027.” Teaser in quiet acknowledgment of an internal delay.
2
Anthropic launched Opus 5.5 , and OpenAI dropped GPT-6 Sol and Luna just minutes after. Sol is now $2/$10 per million tokens, half the price of GPT-5.6. Luna is $0.10/$0.50.
The AI pricing war is now a matter of minutes, not weeks.
Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference more efficient, and we’re passing the savings directly to you: 50% lower API prices for Sol and Luna compared with GPT‑5.6 promotional pricing.
2
Anthropic just launched Claude Opus 5.5: same performance tier as Fable 5.1, at 40% lower cost than Opus 5. Hours later, OpenAI cut GPT-6 Sol and Luna prices in half. The AI price war just got a lot more real 🧵
1
8
Also today: Apple is testing a screenless fitness band to rival Whoop, Alibaba unveiled a custom AI chip aiming to train 10-trillion-parameter Qwen models, and Cognex is acquiring RealSense for $500M to get into robotic perception.
1
3
Here is the full post: apricate.beehiiv.com/p/claud…
1
Anthropic just rolled out Claude Opus 5.5: it matches Fable 5.1 on most tasks at 40% less cost than Opus 5 and 30% faster output. It’s also Anthropic’s first release since CEO Dario Amodei called to “pace the frontier” earlier this month.
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
7
Google just opened preorders for Googlebook without announcing a price. The specs — 2.8K OLED, 14-hour battery, deep Android sync — are real. What you're actually agreeing to pay isn't out yet.
Introducing Googlebook, a new category of laptop, available for preorder today.
💪 Crafted with 2.8K OLED touchscreen displays, 14 hour battery life and all the performance needed to power your ideas
📳 Engineered to sync effortlessly with your @Android phone, so you can jump between your phone and laptop without skipping a beat
✨ Designed for Gemini Intelligence, with personalized, proactive help when and where you need it
14
Grok 4.7 launched at the same $2/$6 per million tokens pricing as 4.6. Coding increased from 40.4% to 46.3% on CursorBench, and terminal work nearly doubled (20.3% to 38.0%).
Still a few benchmarks behind Claude Fable 5.1, but at a fraction of the cost.
6