AI Market Shifts: Anthropic Valuation Debates and New Open Model Releases
Explore the latest AI market updates, including Anthropic's revenue and valuation discussions, Thinking Machines' Inkling 2 Small, and open-weight model releases.
The artificial intelligence sector is navigating a period of shifting valuations and rapid open-weight model introductions. While broader financial markets experience downward corrections, key players in the generative AI space continue to scale their annual run rates and adjust target valuations amid changing economic conditions.
Anthropic's year-end target sits at a $100 billion run rate, with current figures reaching $80 billion as of October. Market observers and investors have debated appropriate IPO valuations, with estimates ranging between $1 trillion and $1.5 trillion against a backdrop of broader tech market adjustments.
New Open Models and Training Efficiency
Recent weeks have brought notable releases in open-weight models. Thinking Machines launched Inkling 2 Small, a natively multimodal Mixture of Experts model featuring 594 billion total parameters and 17 billion active parameters. The model accepts image inputs and supports variable effort, with full weights slated for Apache 2.0 release. Independent evaluations highlight its strong performance on benchmarks like ARC-AGI and Code Arena, alongside observations regarding training efficiency that some experts compare favorably to existing architectures.
Moonshot also released open weights for Kimi K3.1 Pro, a 2.8-trillion-parameter MoE model with 104 billion active parameters, which secured top rankings in Agent Arena. Meanwhile, Baidu introduced the ERNIE-5.5-Flash-Base model, a 280-billion-parameter MoE featuring multi-token prediction and extensive context handling, though its license excludes the EU, UK, and South Korea.
What it means for developers
Developers navigating this fast-paced ecosystem have access to an expanding array of open-weight models, specialized fine-tunes like Nous Research's Arete-27B, and advanced serverless inference solutions. Whether you are building agents, managing long-context workloads, or testing different model capabilities, having flexible access is essential. Developers can try top AI models cheaply through one API at https://apixoai.online, making it easier to integrate and benchmark models from Claude to DeepSeek without managing multiple API keys.
Safety, Evaluation, and Infrastructure
Evaluations continue to stress-test frontier systems. Security and autonomy benchmarks, such as Vals' CyberOffense testing, show that models like GPT-6 Sol and DeepSeek V4.1 Flash can autonomously navigate complex tasks, raising important questions about cybersecurity and infrastructure hardening. Concurrently, new hardware optimizations—such as Zyphra's FP8 attention kernels on AMD hardware and new serverless loop options from providers like Baseten—are driving down inference latency and expanding the boundaries of local and cloud execution.
Source: Anthropic IPO: Whither Now? — Latent Space. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
