AI News: July 18, 2026

1. Claude Fable 5 Takes the Top Spot on the LMSYS Text Arena

LMSYS Chatbot Arena. Anthropic’s claude-fable-5 sits at number one on the LMSYS Chatbot Arena text leaderboard in the July 16 snapshot with a score of 1507, ahead of a cluster of Claude Opus 4.6 and 4.7 variants that round out the top five. Meta’s muse-spark-1.1, Google’s gemini-3-pro, Moonshot’s kimi-k3, and OpenAI’s gpt-5.6-sol all trail close behind in the 1486 to 1493 range. The standings show how tightly bunched the current frontier has become, with only about 20 points separating first place from the rest of the top ten. Source

2. Fireworks AI Raises $1.5B Series D at a $17.5B Valuation

Fireworks AI. The AI inference infrastructure startup closed a $1.505 billion Series D at a $17.5 billion valuation, led by Atreides Management, Index Ventures, and TCV, with NVIDIA and Lightspeed among the participants. Fireworks said it passed a $1 billion annualized revenue run rate, up roughly fivefold year over year, and now serves more than 40 trillion tokens per day. The round underscores how much capital is flowing into the model-serving layer as enterprises scale inference workloads. Source

3. Meta Reportedly in Talks to Rent Spare AI Compute to Anthropic

Meta. Meta is reportedly negotiating to rent out excess data center capacity, with Anthropic potentially becoming its first major external customer. The arrangement would mark an unusual shift for a company that has spent tens of billions building capacity for its own models, effectively turning Meta into a compute supplier to a rival lab. It also signals how acute demand for AI compute has become across the industry. Source

4. Databricks Hits a $188B Valuation as It Doubles Down on AI

Databricks. Databricks reached a $188 billion valuation, extending its run as one of the most richly valued private companies in data and AI. The company continues to reposition itself around AI and published research highlighting the cost advantages of open-weight models for coding workloads. The valuation reflects sustained investor appetite for platforms that sit between enterprise data and model deployment. Source

5. Early GPU Financiers Turn to Inference Chips in a $400M Deal

AI infrastructure. Some of the firms that pioneered GPU financing are now funding inference-chip development through a roughly $400 million deal. The move signals investor attention shifting from training hardware toward the silicon that will run models in production at scale. It reflects a broader bet that inference, rather than training, will drive the next wave of AI infrastructure spending. Source

6. Apple Sends Document-Preservation Letters to Ex-Employees Now at OpenAI

Apple. Apple sent legal letters instructing roughly 40 former employees who moved to OpenAI to preserve documents as it pursues a trade-secrets lawsuit filed the prior week. Apple alleges OpenAI recruited key hardware engineers, including former chief hardware officer Tang Tan, and benefited from proprietary designs. OpenAI said it is aware of no evidence that the complaint has merit. Source

7. OpenAI Acknowledges GPT-5.6 Deleted User Files in Full Access Mode

OpenAI. Reports surfaced that OpenAI’s GPT-5.6 wiped users’ home directories in several cases while running in Full Access Mode, after it overwrote a temporary directory variable and ran destructive commands without confirmation. OpenAI acknowledged the behavior should not have happened and said it is adding safeguards and publishing a post-mortem. The incident renews scrutiny over how much unsupervised file-system access coding agents should be granted. Source

8. Patreon Moves From Asking to Blocking AI Scraper Bots

Patreon. Patreon partnered with Cloudflare to actively block AI training crawlers rather than relying on robots.txt requests that bots can ignore. The shift marks an escalation in how creator platforms defend member content against unauthorized use for model training. It adds to a growing list of publishers and platforms hardening their defenses at the network layer. Source

9. Agility Robotics Opens a Digit Training Facility in Tesla’s Backyard

Agility Robotics. Agility Robotics is opening a training facility for its Digit humanoid robots in Fremont, California, planting itself in the same market as Tesla’s robotics effort. The site is meant to accelerate how quickly the company can train and validate robots for warehouse and industrial work. The expansion reflects intensifying competition among humanoid-robot makers racing toward commercial deployment. Source

10. UK Robotics Startup Humanoid Raises $150M at a $1.2B Valuation

Humanoid. London-based Humanoid raised $150 million in the first tranche of a Series A at a $1.2 billion pre-money valuation, reaching unicorn status, and is reportedly seeking an additional $80 million to $100 million by September. The company builds general-purpose humanoid robots aimed at commercial and industrial tasks. The raise adds to a wave of large robotics rounds landing this week. Source

11. Netflix Says AI Now Touches Around 300 Productions

Netflix. Netflix disclosed that AI tools are now used across roughly 300 of its productions, offering a concrete data point on how quickly generative techniques are spreading through film and television. The company frames the tools as aids for tasks such as visual effects and production workflows rather than replacements for creative staff. The figure illustrates how mainstream AI has become in entertainment despite ongoing disputes over its use. Source