AI Daily Newsletter: Preview of GPT-5.6 Sol and OpenAI's Self-Designed Jalapeño Chip

🔥 1. GPT-5.6 Sol Officially Previewed — US Government to Control Who Gets to Use It

OpenAI has announced the GPT-5.6 line with three versions: Sol (flagship), Terra (balanced), and Luna (fast, affordable). The most notable point is that the US government has requested OpenAI to approve each customer before using Sol, causing public concern about a de facto AI licensing mechanism. OpenAI stated that they do not want this to become the “default” for future releases. Sol achieves SOTA on Terminal-Bench 2.1 and GeneBench v1 with pricing at $5/$30 per 1M token.

Source: OpenAI Blog | Washington Post | TechCrunch

🤖 2. Anthropic Allowed by US Government to Release Claude Mythos 5

The US Department of Commerce has officially allowed Anthropic to release Claude Mythos 5 — their most powerful model — to over 100 trusted partner companies and organizations, including many Fortune 500 names. This is the result of two weeks of Anthropic’s cooperation with the government to address security risks. This move shows the trend of the US government increasingly intervening in the control of frontier AI models.

Source: Reuters

💾 3. OpenAI Launches Self-Designed AI Chip “Jalapeño” — A Step Away from Nvidia

OpenAI has officially announced its first AI chip designed in-house, manufactured in collaboration with Broadcom. This move is part of the strategy to reduce dependence on Nvidia, which currently dominates the AI chip market. Not only OpenAI but also SpaceX, Tesla, and many other big tech companies are building their own chips, putting significant pressure on Nvidia.

Source: TechCrunch

💰 4. AI Startup Lindy Ditches Claude for DeepSeek Due to “Unsustainable” Costs

Lindy, a 25-employee AI startup, has completely switched from Anthropic Claude to DeepSeek (hosted in the US). CEO Flo Crivello stated directly: “This is a matter of the company’s survival.” AI costs have surpassed personnel costs, and the switch helps them save millions of dollars. Altman has also acknowledged that AI costs are a “big problem” when agentic systems consume enormous tokens.

Source: The Decoder

🎮 5. General Intuition Raises $2.3B: Using Games to Train AI Agents for the Real World

General Intuition has successfully raised $2.3 billion with the ambition of using video games as an environment to train AI agents that can operate in the real world. This idea is not new, but the scale of investment shows that VCs are betting big on “embodied AI” — where AI not only knows how to chat but also how to act.

Source: TechCrunch