In partnership with

Your employees are connecting AI to everything. Now what?

ChatGPT and Claude don't just answer questions anymore. Employees are connecting them directly to Notion, Linear, Jira, and the rest of your stack. The AI can read, write, and take actions on company data. Most IT and security teams have no visibility into any of it.

Harmonic Security Connectors changes that. It sits inline with every AI-to-app connection, so you see each call, control what data moves, and block destructive actions before they happen. Employees notice nothing different.

See what's actually running across your business in a live demo.

See what's connected

OpenAI and Anthropic cut costs 90 minutes apart. Plus Google, Meta, Japan and Digger.
NEON DREAMS AI
NEON DAILY
SEP 23, 2026
AI    TECHNOLOGY    BUSINESS    WORLD    CULTURE
TOP STORY  •  AI & MODELS  •  Anthropic / OpenAI

The AI Price War Arrives: OpenAI and Anthropic Cut Costs on the Same Afternoon

Two rival launches, about 90 minutes apart, with the same pitch: strong models that are cheaper to run.

On Tuesday, the two biggest names in AI shipped new models within roughly an hour and a half of each other. Neither led with a leap in raw intelligence. Both led with price.

Anthropic went first with Claude Opus 5.5. The company says it performs at the level of Claude Fable 5.1 on most work and costs about 40% less to run than Opus 5 on typical workloads. Only part of that comes from the price tag: per-token rates fell 20%, to $4 per million input tokens and $20 per million output tokens. The rest, Anthropic says, comes from the model using fewer tokens to finish a job. Cache reads, a big share of costs for coding and agent work, dropped 60%.

OpenAI followed with GPT-6 Sol and GPT-6 Luna, two models that sit below its flagship, GPT-6 Astra. Sol is priced at $2 per million input tokens and $10 per million output tokens; Luna at $0.10 and $0.50. OpenAI says both cost half of what their GPT-5.6 predecessors charge under current promotional pricing.

Why it matters: For businesses and developers that call these models thousands or millions of times a day, small per-token differences add up quickly. Cheaper capable models make more tasks worth automating and more experiments worth running.

Two different plays: OpenAI is spreading its lineup across very different price points, from a premium flagship down to a budget model for high-volume work. Anthropic is keeping one top-tier model and making it cheaper to operate.

The catch: The launch-day numbers are the companies' own. Both increasingly compare cost per task rather than cost per token, which is harder for outsiders to check. Real-world bills will depend on the job, reasoning settings, caching and how much output an app actually needs.

The bigger picture: Pressure is coming from below, as low-cost open-weight models from Chinese developers such as Alibaba and DeepSeek handle more everyday work, according to The Next Web. The next phase of AI competition looks less like a race for the smartest model and more like a race to make smart models affordable at scale.

READ THE SOURCE
MORE TOP STORIES
02
TOOLS & BUILDS  •  Meta Engineering

Meta Open-Sources the Tool That Decides Where Its Servers Go

Meta has released Rebalancer, the software it has used for more than nine years to solve assignment problems: which racks go in which data centers, which servers serve which services, and where user traffic gets routed. Meta says it now solves roughly 40 million of these problems a day across more than 30 problem types.

Users describe the problem in plain building blocks (objects, bins, rules and goals), and Rebalancer picks how to solve it: a fast local-search method for huge problems, or an exact optimization approach for smaller ones. It ships under the Apache 2.0 license with a visual debugging tool.

Why it matters: It's a rare look at the plumbing behind a hyperscale company, and the same math applies well beyond tech, from logistics to scheduling.
Read the source →
03
AI SAFETY  •  The Next Web

OpenAI Spells Out What Outside Safety Reviewers Should Test

OpenAI has published guidance on how independent third parties should assess its frontier models. It names four areas for review: the company's safety cases, its safeguards against misuse, its dangerous-capability evaluations, and investigations of incidents where models misbehave. It also lays out seven principles on access, independence, conflicts of interest and publication.

OpenAI says outside assessors should be able to challenge its assumptions and reach their own conclusions about whether its safeguards work.

The catch: Some terms favor the lab. Companies would get time to fix problems before findings go public and could request redactions, though assessors keep editorial control and may note where material was cut. OpenAI hasn't named which groups will do the work.
Read the source →
04
BUSINESS & POWER  •  U.S. Dept. of Justice / AdExchanger

Google Keeps Its Ad Business, but Must Open It to Rivals

The full 106-page remedies ruling in the U.S. ad-tech antitrust case is now public, filling in the details behind Judge Leonie Brinkema's Sept. 2 decision not to break up Google's advertising business.

Instead of forcing a sale of its AdX exchange, the court ordered Google to connect AdX and its DFP ad server to the open-source Prebid system, send AdX bids to rival ad servers, let publishers access and export their own data, and stop its AdWords tool from favoring Google's own ad products. A court-appointed monitor will oversee compliance for six years. The two sides must file a joint proposed final judgment by Oct. 2.

Why it matters: It's the second time in about a year a federal judge has declined to break up Google, choosing rules over a split. Whether that works now depends on enforcement.
Read the source →
05
WORLD  •  Prime Minister's Office of Japan

Japan Plans to Host a Global AI Summit

Speaking at the UN General Assembly on Tuesday, Prime Minister Sanae Takaichi said Japan intends to host an AI Summit at the earliest opportunity. No date or city has been announced.

Takaichi tied the plan to Japan's Hiroshima AI Process, which produced the first international principles and code of conduct for generative AI, and to work with partners on a safe, secure and trustworthy AI ecosystem, including cooperation on the risks of highly capable frontier models.

Read the source →
06
CULTURE  •  TheWrap

First Reactions to Tom Cruise's 'Digger' Are Split

The social embargo lifted Tuesday on Alejandro G. Iñárritu's new Tom Cruise film, and the early verdict is mixed. Critics broadly praised Cruise's fully committed performance, the supporting cast, including Riz Ahmed and John Goodman, and the film's production design. The satire and tone drew the most complaints.

Variety's Clayton Davis summed it up as a big swing that doesn't fully connect and predicted a wide divide among audiences. The Warner Bros. film opens in U.S. theaters Oct. 2.

Read the source →
NEON DAILY
NEON DREAMS AI • AI • TECHNOLOGY • CULTURE
Six stories, fact-checked and edited for September 23, 2026. Not for redistribution without attribution.

Most teams add a second database for analytics. Then manage sync, lag, and drift forever. TimescaleDB extends Postgres instead. Hypertables, 95% compression, aggregates. No pipeline.

Get $1000 Credit

Reply

Avatar

or to participate

Keep Reading

View more
caret-right
© 2026 Neon Newsletter.
beehiivPowered by beehiiv