OpenAI:

We’re releasing gpt-oss-120b and gpt-oss-20b — two state-of-the-art open-weight language models that deliver strong real-world performance at low cost. Available under the flexible Apache 2.0 license, these models outperform similarly sized open models on reasoning tasks, demonstrate strong tool use capabilities, and are optimized for efficient deployment on consumer hardware. They were trained using a mix of reinforcement learning and techniques informed by OpenAI’s most advanced internal models, including o3 and other frontier systems.

The gpt-oss-120b model achieves near-parity with OpenAI o4-mini on core reasoning benchmarks, while running efficiently on a single 80 GB GPU. The gpt-oss-20b model delivers similar results to OpenAI o3‑mini on common benchmarks and can run on edge devices with just 16 GB of memory, making it ideal for on-device use cases, local inference, or rapid iteration without costly infrastructure. Both models also perform strongly on tool use, few-shot function calling, CoT reasoning (as seen in results on the Tau-Bench agentic evaluation suite) and HealthBench (even outperforming proprietary models like OpenAI o1 and GPT‑4o).

Simon Willison:

The long promised OpenAI open weight models are here, and they are very impressive. [...]

o4-mini and o3-mini are really good proprietary models — I was not expecting the open weights releases to be anywhere near that class, especially given their small sizes. That gpt-oss-20b model should run quite comfortably on a Mac laptop with 32GB of RAM.

Anthropic:

GitHub notes that Claude Opus 4.1 improves across most capabilities relative to Opus 4, with particularly notable performance gains in multi-file code refactoring. Rakuten Group finds that Opus 4.1 excels at pinpointing exact corrections within large codebases without making unnecessary adjustments or introducing bugs, with their team preferring this precision for everyday debugging tasks. Windsurf reports Opus 4.1 delivers a one standard deviation improvement over Opus 4 on their junior developer benchmark, showing roughly the same performance leap as the jump from Sonnet 3.7 to Sonnet 4.

Nothing spectacular here, but incremental improvements add up. Mike Krieger — best known as a co-founder of Instagram, now chief product officer at Anthropic — in an interview with Bloomberg:

“In the past, we were too focused on only shipping the really big upgrades,” said Anthropic Chief Product Officer Mike Krieger. “It’s better at coding, better at reasoning, better at agentic tasks. We’re just making it better for people.” [...]

“One thing I’ve learned, especially in AI as it’s moving quickly, is that we can focus on what we have — and what other folks are going to do is ultimately up to them,” Krieger said when asked about OpenAI’s upcoming release. “We’ll see what ends up happening on the OpenAI side, but for us, we really just focused on what can we deliver for the customers we have.”

I’m on board with the idea that Apple need not acquire any of these AI startups, but if they do, Anthropic — not Perplexity — seems the one most aligned with Apple’s values. And I don’t mean values in just an ethical sense, but their entire approach to product development in general.

Tom Warren:

In a new Pixel 10 ad, Google dunks on Apple’s failed promise of Siri AI improvements, with a narrator that suggests you could “just change your phone” if you bought “a new phone because of a feature that’s coming soon, but it’s been coming soon for a full year.”

The 30-second spot appeared on YouTube and X today, teasing the launch of Google’s new Pixel 10 devices on August 20th.

The whole Siri/Apple Intelligence thing has been an enormous self-inflicted embarrassment, but when it comes to Pixel phones, all I can think of is that Mad Men “I don’t think about you at all” GIF.

Ashley Belanger, writing for Ars Technica:

Porn sites may have blown up Meta’s key defense in a copyright fight with book authors who earlier this year said that Meta torrented “at least 81.7 terabytes of data across multiple shadow libraries” to train its AI models. [...]

After authors revealed Meta’s torrenting, Strike 3 Holdings checked its proprietary BitTorrent-tracking tools designed to detect infringement of its videos and alleged that the company found evidence that Meta has been torrenting and seeding its copyrighted content for years — since at least 2018. Some of the IP addresses were clearly registered to Meta, while others appeared to be “hidden,” and at least one was linked to a Meta employee, the filing said.

According to Strike 3 Holdings, Meta “willfully and intentionally” infringed “at least 2,396 movies” as part of a strategy to download terabytes of data as fast as possible by seeding popular high-quality porn. Supposedly, Meta continued seeding the content “sometimes for days, weeks, or even months” after downloading them, and these movies may also have been secretly used to train Meta’s AI models, Strike 3 Holdings alleged.

The porn site operator explained to the court that BitTorrent’s protocol establishes a “tit-for-tat” mechanism that “rewards users who distribute the most desired content.” It alleged that Meta took advantage of this system by “often” pirating adult videos that are “often within the most infringed files on BitTorrent websites” on “the very same day the motion pictures are released.”

Meta is an empty husk of a company with no values, no beliefs, other than growth and dominance for the sake of growth and dominance.

Ghost:

When we announced Ghost 5.0 a few years ago, we were proud to share that Ghost’s revenue had hit $4M — while publisher earnings had surpassed $10M. It felt great to have such a clear sign that our goal to create a sustainable business model for independent creators was succeeding.

Today, Ghost’s annual revenue is over $8.5M while total publisher earnings on Ghost have now surpassed $100M. [...]

Unlike our venture-backed peers obsessed with growth at all costs, we’re structured as a non-profit foundation that serves publishers directly with open source software. We believe independent media cannot be beholden to proprietary tech companies, so Ghost publishers don’t just “own their email list” — they own the entire software stack that underpins their business, end to end.

Not a centralized platform controlled by a single corporation, but open infrastructure that’s shared by everyone.

Aside from my feelings about Substack — clearly the main target of Ghost’s shade-throwing here — it’s just great to see so many indie publishers and writers thriving on Ghost.