Anthropic abruptly removed its Fable 5 model from all markets under a US government national security order, marking the first known instance of a state forcing a model's withdrawal. The takedown prompted warnings about closed-source AI becoming a restricted commodity, with open-source models now seen as critical. Kimi released its K2.7 Code model, achieving a 20% coding ability improvement over K2.6 and reducing thinking tokens by 30%, with API pricing at $6.5/m input and $27/m output. In response, Zhipu announced it will soon release GLM 5.2, emphasizing that frontier intelligence should remain open and available to all developers.
A commentary claims that models with the largest market share are more prone to hallucination because cost pressures limit investment in measures to reduce false outputs. It notes that abroad, many users rely on free GPT models and consequently obtain responses with very high hallucination rates, attributing this to the cost-driven design of freely available models.
Baidu’s autonomous ride‑hailing service Apollo Go (Luobo Kuaipao) operated in a dense urban area of Wuhan, driving from Optics Valley to Caida Snack Street. During the trip the vehicle encountered a sudden cut‑in by another car but managed to avoid a collision completely. The incident demonstrates the system’s capability to handle aggressive urban driving scenarios without human intervention.
A user experimented with Doubao’s Seedance 2.0 video generation model and found that converting a real person’s photo into a painted style, then using that as a reference image with a prompt specifying real-photo style, allows the model to generate realistic human faces while bypassing its face review mechanism. The output, paired with a French romantic style, achieved a high-end brand advertising feel. The user shared the method and will provide the exact prompt later.
Claude Fable 5 will be removed from the Token Plan after June 22, leaving only API access. Its safety guardrails block even basic biology questions like 'What is mitochondria?', requiring users to apply for the Mythos plan for biopharma research. Anthropic also prohibits using Fable for model-related R&D, with the official rationale that models can accelerate their own development, effectively barring competitors from improving their own models with it.
Claude Fable 5 is released as the most significant model advancement since Claude 4.5, described as the best widely available model. It is built on a Mythos base with enhanced safety guardrails and praised by Andrej Kapathy. The model achieves top-tier performance across software engineering, knowledge work, vision, and scientific reasoning, especially excelling on complex, long-horizon tasks. Pricing is set at $10 per million input tokens, $50 per million output tokens, and $1 for cached input. The model will be available on Cola at its original price.