This brief covers AI news from 2026-09-12 UTC.
Anthropic’s new Claude pair lands with a price cut, Cognition brings Fusion to Devin, and Nvidia circles Anthropic’s IPO.
Anthropic ships Claude Fable 5.1 and Mythos 5.1
What happened: Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1, calling them its most advanced models for coding and knowledge work. Fable 5.1 is generally available and costs an estimated 25% less than Fable 5 for typical token-billed workloads, with savings up to about 45% on highly agentic work.
Why it matters: Cheaper cache reads and fewer false-positive blocks in cybersecurity lower the cost and friction of running agentic coding and security workflows on Claude.
Cognition brings Fusion to Devin Desktop and CLI
What happened: Cognition launched Fusion in Devin Desktop and CLI, a harness that pairs a frontier model for planning and review with a cheaper model for execution. It reports up to 39% better efficiency than other harnesses across major coding benchmarks, recommending Fable 5.1 with SWE-2.
Why it matters: Developers can cut coding agent costs by routing execution to a cheaper model while keeping frontier planning, with reported savings of 11% to 46% across benchmarks.
Nvidia in talks to anchor Anthropic’s IPO with up to $10B
What happened: Reuters reported that Anthropic is in talks to bring Nvidia in as an anchor investor for an IPO seeking up to $100 billion at a valuation of around $2 trillion. Nvidia is considering investing up to $10 billion, sources said, with plans still under discussion.
Why it matters: An Nvidia anchor stake would deepen ties between the chipmaker and a major customer, shaping compute supply and pricing for teams building on Claude.
Cohere in talks for up to $3B raise at $20B valuation
What happened: Bloomberg reported that AI startup Cohere may secure up to $3 billion in funding, in a round that could value the company at $20 billion.
Why it matters: A larger Cohere war chest means more competition in enterprise and sovereign AI deployments, an alternative to the largest US labs for builders with data residency needs.
Positron AI raises $875M at $5B for inference silicon
What happened: Positron AI announced an $875 million Series C at a $5 billion post-money valuation, co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital, SemiAnalysis Capital and Jim Clark. It builds memory-first inference systems on commodity LPDDR5X memory.
Why it matters: More capital for inference-specific silicon could ease memory bandwidth and power constraints that currently limit tokens per dollar for serving models at scale.
d-Matrix plugs Raptor accelerators into Nvidia NVLink Fusion
What happened: AI inference chipmaker d-Matrix signed a multi-year deal to bring its forthcoming Raptor XPUs into Nvidia’s NVLink Fusion ecosystem and MGX rack architecture, alongside Vera CPUs, NVLink switches and BlueField-4 DPUs. Raptor is expected to tape out before the end of 2026.
Why it matters: Rivals can now slot custom inference silicon into Nvidia racks, giving operators a lower-risk path to ultralow-latency decode for coding assistants and voice agents.
Mecka AI nears $500M valuation in Sequoia-led round
What happened: TechCrunch reported that Mecka AI, a startup collecting human motion data to train humanoid robots, is nearing a round led by Sequoia Capital at a valuation of about $500 million. The round comes three months after Mecka raised $60 million led by Framework Ventures.
Why it matters: Robot training data is becoming a funded category, a signal that physical-world datasets may follow the trajectory of LLM training data vendors.