Lilian Weng, a co-founder of Thinking Machines Lab and former OpenAI vice president of research and safety, said on social media that tomorrow will be her last day at the company. Her departure comes 20 months after she co-founded the startup with former OpenAI CTO Mira Murati, and only days after Thinking Machines released its first open-source large language model, Inkling. In her departure note, Weng said health had become the central issue. She wrote that over the past seven months she had been sick more often than at any other point in her life, including during the company’s push to launch Inkling, when she had to take sick leave and struggled with guilt over stepping away from work. She said she had considered taking on a smaller role with less pressure and uncertainty, but concluded that such an arrangement did not fit how she works. The move also lands against a more complicated backdrop for Thinking Machines. The company raised about $2 billion in a seed round in July 2025 at a $12 billion valuation, then later faced scrutiny over valuation, product pace, compute spending and staff turnover. Even so, it has launched its Tinker fine-tuning tool, released Inkling, signed a multibillion-dollar compute agreement with Google Cloud and reached a multiyear strategic partnership with Nvidia.

Crypto markets head into the week of July 27 with a dense calendar of macro releases, central bank decisions and industry-specific events that could shape digital-asset pricing. According to CoinDesk’s weekly preview, the Federal Reserve, Bank of England and Bank of Japan are all expected to leave rates unchanged, but traders are watching closely for any signal that higher energy prices could push policymakers closer to renewed tightening. CME FedWatch puts the odds of a U.S. rate hike at 33%, while prediction markets imply 19%, up from near zero earlier in the month. Thursday brings a heavier test with second-quarter U.S. GDP and June Personal Consumption Expenditures data. CoinDesk said a combination of firm growth and sticky inflation could strengthen the higher-for-longer rates trade through higher yields and a stronger dollar, while softer numbers could unwind that setup. On the crypto side, Polygon is set to deploy its Ithaca hard fork on mainnet, BitMEX will settle and delist 35 derivatives contracts, and FTX Recovery Trust is scheduled to begin its fifth creditor distribution, worth roughly $900 million. Earnings from Robinhood, Coinbase and Strategy are also due, alongside governance votes, token unlocks and industry conferences.

Thinking Machines Lab released its open-source model Inkling on July 15. The model has 975 billion total parameters, uses a mixture-of-experts architecture, and supports text, image, and audio processing. It also offers a context window of up to 1 million tokens. Inkling has been open-sourced on Hugging Face under the Apache 2.0 license. In MCP Atlas agent task testing, it scored 74.1%. The model is also available through OpenRouter, where input pricing is set at $1 per million tokens. The item was cited by Techub, with Decrypt named as the source in the brief.

Mira Murati’s Thinking Machines Lab has released Inkling, its first model after two years of work following her departure from OpenAI. Decrypt’s review frames it as the strongest open-source model trained from scratch by a Western lab, with full weights on Hugging Face under Apache 2.0, a 975 billion-parameter mixture-of-experts architecture, 41 billion active parameters at inference, multimodal input support, a 1 million-token context window, and pretraining on 45 trillion tokens. The model scored 74.1% on MCP Atlas and 77.6% on SWE-Bench Verified, outperforming Nvidia’s Nemotron on both tests, and is available on OpenRouter at $1 per million input tokens and $4.05 per million output tokens. The review becomes less favorable in direct usage. In coding tests, Inkling failed a demanding 1,955-word prompt and produced only a limited result on a simplified 99-word version, while a 27B Bonsai model running on a phone delivered a more complete game from the same prompt. Decrypt also found weaknesses in logic, associative creativity, and creative writing, and described the model as heavily censored after it refused both of the publication’s safety prompts. Its conclusion is narrow rather than sweeping: Inkling looks most useful for compliance-focused organizations that want an open, modifiable Western model, but it is not yet the obvious daily choice for developers optimizing for coding quality, uncensored output, or price-performance.

IOSG’s latest deep dive argues that Robinhood is no longer renting blockspace from someone else’s network. By launching its own Layer 2, the company now controls the rails for trading, settlement, collateral, yield and asset circulation, a move the report frames as a direct answer to Coinbase’s Base. Robinhood Chain was built for tokenized equities and broader real-world asset activity, yet its first wave of traction came from meme coins and AI agents rather than stocks. As of July 20, 2026, IOSG said RWA assets still accounted for only about 4% of total value locked. The report lays out a layered structure around the chain, including USDG as the main settlement dollar, USDe as a yield-bearing collateral asset, Wallet as the user entry point, and separate perpetual venues such as Lighter and Arcus. It also spends considerable time on the legal structure of Robinhood’s stock tokens, describing them as tokenized debt securities issued by Robinhood Assets (Jersey) Limited rather than direct ownership of underlying shares. IOSG’s conclusion is that the infrastructure economics already make sense for Robinhood, but the unresolved question is whether meme-driven traffic can be converted into durable RWA activity and whether the company will eventually publish reserve proof for its stock token model.

Chainlink expands CCIP to Creditcoin, Neo X, and Tempo, integrates CRE/Data Feeds on Ink, and migrates data streams to Robinhood Chain testnet, boosting cross-chain interoperability for tokenized assets and DeFi.

Binance has listed Ripple's USD-pegged stablecoin RLUSD, offering XRP and dollar pairs with margin eligibility and upcoming Earn integration. Ripple reveals reserves exceeding 103% in T-bills and FDIC-insured deposits, and plans to expand to multiple Ethereum L2 networks in 2026 via Wormhole and NTT. NGOs including World Central Kitchen and Water.org are testing RLUSD for humanitarian aid payments.

David Schwartz said XRPL is moving beyond payments into tokenized real-world assets. As RLUSD expands across chains, activity on XRPL, stablecoin usage, and tokenized asset value have all climbed.