X Open-Sources Codebase; Grok 4.5 Tops Benchmarks

Elon Musk announces X will open-source its entire codebase, inviting third-party audits to verify consistency.
Sam Altman reports explosive growth for 5.6 Sol and 8M active users across Codex and ChatGPT Work, with rate limits removed.
Grok 4.5 tops FrontierSWE and Long-Horizon Terminal-Bench, while GPT-5.6 scores IQ 136 and solves a major statistical problem.
Perplexity open-sources WANDR benchmark for depth/breadth research; Vercel releases token-flow dataset for AI Gateway.
Greylock launches $1.5B early-stage fund (Greylock 18) focused on AI; Boom Supersonic survives near-shutdown with investor support.
Garry Tan identifies Vorflux AI coding agent as a potential $100B company; AI agents gain phone-calling ability via Claude Code.

Elon Musk announces that after safety review, X will open-source its entire codebase, with third-party audits to ensure running code matches open-source version. — via 1
Sam Altman reports that 5.6 Sol is growing rapidly, with the inference team scaling to meet demand; Codex and ChatGPT Work have 8 million active users, reset usage limits, and removed 5-hour rate limits to encourage exploration of GPT-5.6 Sol's capabilities. — via 1 2
Elon Musk asserts Grok 4.5 ranks top on FrontierSWE and Long-Horizon Terminal-Bench, outperforming Claude Opus and GPT series with low cost and fast long-horizon task processing; it adds Stripe and Calendly connectors, positioning as an operating system for workflows. — via 1 2 3
Garry Tan highlights GPT-5.6's major advances: initial IQ score of 136 (exceeding 99% of humans) and solving a key open problem in statistics (paper cited >130k times). — via 1 2
Aravind Srinivas announces Perplexity open-sources WANDR, an internal benchmark for deep and broad research with 500 tasks based on real knowledge work, where even the strongest models struggle. — via 1 2

Elon Musk announces Starlink V5 terminal: 48% smaller, 62% lighter, lower power, higher wind rating, supports 375+ Mbps, designed and made in USA. — via 1 2
Guillermo Rauch releases open token-flow dataset for Vercel AI Gateway, containing real production usage data for download and query. — via 1
Elon Musk reports SpaceX launched two Falcon 9 rockets from two pads in under 8 hours, deploying 56 Starlink satellites; Starship remains on schedule for tomorrow's flight. — via 1 2
Yohei highlights that the US Medicare claims system runs on 18 million lines of assembly and COBOL code with 50 years of policy logic; a team documented over 100k business rules to help policymakers understand the system. — via 1

Reid Hoffman announces Greylock's new $1.5B early-stage fund (Greylock 18) focused on founders who anticipate the future, stating that AI is reinventing every sector and defining AI companies are yet to be born. — via 1
Reid Hoffman shares that Boom Supersonic faced a board request to shut down with only weeks of cash left, but survived with support from investors including Hoffman, making supersonic flight viable. — via 1
Garry Tan invests in Vorflux AI coding agent, calling it far superior to Devin, predicting it will become a $100B+ company. — via 1 2
Ryan Hoover highlights a robotics fulfillment company that raised $13M, enabling e-commerce brands to reduce costs by 80% and improve delivery speed. — via 1

Garry Tan details an approach to building AI long-term memory systems using Karpathy's LLM Wiki pattern, GBrain graph retrieval, and Hermes agent layer, emphasizing building human-readable knowledge sources first. — via 1
Yohei notes that Claude Code can now make phone calls, enabling agents to book restaurants and track leads in 15 seconds, powered by @trysaperly and @monid_ai. — via 1
Guillermo Rauch showcases Vercel Agent reducing build times by 10x based on user feedback, and Web Analytics API allowing agents to correlate visitors with custom events and deployment changes. — via 1 2
Jason Fried criticizes AI customer service, stating that AI bots will make users hate brands, and recommends Basecamp's alternative approach. — via 1

Discover