OpenAI's latest flagship hits general availability in three sizes — Luna, Terra, and Sol — with pricing that undercuts both Claude Opus and Claude Fable 5 at comparable performance tiers. Sol tops out at $5/$30 per million tokens versus Fable 5's $10/$50, a meaningful gap that will pressure competitors on cost. The model is already being benchmarked against Grok 4.5 across the builder community.
Patreon has partnered with Cloudflare to block AI training crawlers, with CEO Jack Conte drawing a hard line on consent and compensation. It's one of the more aggressive platform-level responses to unauthorized scraping. Expect others to follow.
New research finds AI-generated fiction has distinct tells: ChatGPT leans on dream sequences, Gemini over-describes characters. Detection is less about watermarks and more about patterns of aesthetic failure. The models are consistent in their mediocrity.
Meta's Muse Spark 1.1 arrives with the first public API for the Spark model family, claiming significant improvements in agentic tool calling and computer use. It enters a crowded field but Meta's distribution advantage shouldn't be dismissed. The API opening is the real story here.
The FTC's new settlement with John Deere has actual enforcement teeth, unlike prior agreements that changed little. Farmers can now repair their own equipment without voiding warranties or being locked out by proprietary software. A quiet but significant win for hardware ownership rights.
The crypto IPO pipeline is drying up as institutional capital shifts toward AI plays and geopolitical uncertainty keeps risk appetite suppressed. Bitcoin sits flat at $64K while equities grind higher. The rotation story is real and accelerating.
The Kobeissi Letter broke the escalating Strait of Hormuz situation in real time — Iranian missile strike on a commercial vessel, US military response, Iran's parliament declaring "the era of one-sided deals is over." Markets barely flinched: VIX dropped 5%, equities ticked up. Either traders are priced for resolution or dangerously complacent about a chokepoint that moves 20% of global oil.
Sebastian Raschka updated his model benchmark comparison to include Grok 4.5 and Muse Spark 1.1, concluding Grok 4.5 sits at the Pareto frontier — best performance-per-dollar in the current landscape. This kind of practitioner-level benchmark carries more weight than official leaderboards right now, and the community is treating it accordingly.
Ethan Mollick handed Fable 5 a simple game and told it to "do something incredible." It built an entirely new game called DEEP TIME. It's a useful demo of how the best use of frontier models right now isn't task completion — it's creative divergence from a prompt. The gap between "do this" and "surprise me" is where things get interesting.
Logan Kilpatrick announced free custom pretty URLs for deployed apps in Google AI Studio — each app gets its own *.ai.studio domain at no cost. Small feature, meaningful signal: Google is competing hard for the developer layer and lowering friction at every step of the build-and-share loop.
The model wars just got a pricing floor. GPT-5.6 Sol at $5/$30 per million tokens is OpenAI playing offense against Anthropic, and the timing — alongside Grok 4.5 landing at what benchmarkers are calling the Pareto frontier — means we're in a genuine three-way shootout. Builders are running live comparisons in public, and the consensus is forming fast: cost curves are collapsing faster than anyone expected six months ago.
Meanwhile the Hormuz situation is a stress test the market is choosing to ignore. Crypto flat, VIX down, equities up — on a day the US military conducted strikes in a global oil chokepoint. Either this de-escalates fast or the complacency reads as a setup. Worth watching Monday open closely. The rotation from crypto to AI capital continues either way; geopolitics just adds a variable no model is pricing cleanly.