Models
Moonshot AI releases Kimi K3 open weights and infrastructure after shaking up the frontier model race
Moonshot AI just released the open weights and infrastructure for Kimi K3. This move shakes up the frontier model race, allowing developers to build on top of their advancements more easily.
Microsoft launches its own cybersecurity model MAI-Cyber-1-Flash but still depends on OpenAI for the toughest tasks
Microsoft just launched its cybersecurity model MAI-Cyber-1-Flash. It still relies on OpenAI for handling the most complex tasks, indicating a collaborative approach in tackling cybersecurity challenges.
Google is working on a new AI chip designed to make Gemini more efficient
Google is developing a new AI chip to enhance the efficiency of its Gemini model. This upgrade aims to improve performance and reduce energy consumption for AI tasks.
Alibaba's Qwen takes on Kimi K3 with open-weight Qwen 3.8, says model is "second only to Fable 5"
Alibaba just launched the open-weight Qwen 3.8 model, claiming it's 'second only to Fable 5'. This upgrade enhances their competitive edge in the AI model landscape.
Moonshot's Kimi K3 outperforms Fable 5 in frontend code but lags far behind in complex math
Moonshot's Kimi K3 outperforms Fable 5 in generating frontend code but struggles with complex math tasks. Users can expect better performance in web development, but may face limitations in advanced calculations.
Open-weight models now match frontier cyber performance from just four months ago at a fraction of the cost
Open-weight models are now achieving frontier cyber performance that was only available a few months ago, and they do it at a much lower cost. This shift makes advanced capabilities more accessible for various applications in cybersecurity.
[AINews] Kimi K3 2.8T-A50B: the largest open model ever released; Opus 4.8-class at Sonnet 5 pricing
Kimi just released the K3 2.8T-A50B, the largest open model to date, priced competitively at Sonnet 5 levels. This gives developers access to powerful capabilities without breaking the bank.
Kimi's open model K3 nears GPT-5.6 Sol and Fable 5 while signaling the end of super cheap Chinese AI
Kimi is releasing its open model K3, which approaches the capabilities of GPT-5.6 and Fable 5. This shift indicates a potential end to the era of extremely low-cost Chinese AI models, impacting the competitive landscape.
Introducing Open-Weight Models to Claygent - The GTM with Clay Blog
Claygent just introduced open-weight models to their platform. This allows developers to customize and fine-tune AI models for specific applications, enhancing flexibility and usability.
Inkling model from Thinking Machines Lab now on Databricks
Thinking Machines Lab just launched the Inkling model on Databricks. This integration allows users to leverage advanced AI capabilities directly within the Databricks platform for enhanced data processing and analysis.
Thinking Machines Lab Drops Its First Model
Thinking Machines Lab just released its first AI model. This launch marks their entry into the competitive AI landscape, potentially offering new capabilities to developers and businesses.
Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling
Thinking Machines just launched Inkling, its first open model designed to challenge one-size-fits-all AI. This move allows developers to customize AI solutions more effectively for specific tasks and industries.
[AINews] Codex usage up >10x in 6 months to 7M users, +1M in the past ~day; did Codex overtake Claude Code??
Codex usage just surged over 10x in six months, reaching 7 million users with an additional million in just one day. This spike indicates growing reliance on Codex for coding tasks, potentially surpassing Claude Code in popularity.
OpenAI GPT-5.6 Sol, Terra, and Luna are now generally available on Amazon Bedrock
OpenAI just made GPT-5.6 Sol, Terra, and Luna available on Amazon Bedrock. Users can now access these advanced models for various applications directly through the AWS platform.
OpenAI's GPT-5.6 Sol Ultra reportedly solves a 50-year-old math problem in under an hour
OpenAI's GPT-5.6 Sol Ultra just solved a 50-year-old math problem in under an hour. This breakthrough showcases the model's advanced problem-solving capabilities and could impact fields relying on complex mathematical solutions.
Meta's Muse Spark 1.1 outperforms GLM-5.2 in coding and costs slightly less
Meta just released Muse Spark 1.1, which outperforms GLM-5.2 in coding tasks while being slightly cheaper. This upgrade means developers can access a more efficient coding model without breaking the bank.
OpenAI staffer maps out which of GPT-5.6 Sol's five reasoning levels fits which task complexity
OpenAI staffer details how GPT-5.6 Sol's five reasoning levels align with different task complexities. This mapping helps users understand which reasoning level to apply for specific tasks, enhancing usability.
OpenAI launches its new family of models with GPT-5.6
OpenAI just launched its new family of models with GPT-5.6. This update enhances performance and capabilities, making it more effective for developers and users alike.
The new GPT-5.6 family: Luna, Terra, Sol
OpenAI just launched the GPT-5.6 family, introducing models named Luna, Terra, and Sol. These updates enhance performance and usability for developers building AI applications.
GPT-5.6 Sol nearly matches Fable 5 on aggregated benchmarks at one-third the cost
GPT-5.6 Sol just achieved benchmark performance nearly equal to Fable 5 while costing only a third of the price. This makes advanced AI capabilities more accessible for developers and businesses looking to optimize costs.
SpaceXAI releases Grok 4.5, which Elon describes as an ‘Opus-class model’
SpaceXAI just released Grok 4.5, which Elon Musk calls an ‘Opus-class model’. This upgrade enhances AI capabilities for various applications, pushing the boundaries of what's possible in AI technology.
Cohere Transcribe Arabic is an open-source model built for Arabic's toughest transcription problems
Cohere just launched Transcribe Arabic, an open-source model targeting complex Arabic transcription challenges. This model aims to improve accuracy and accessibility for Arabic speakers in transcription tasks.
tencent/Hy3
Tencent just launched Hy3, an AI model designed for generating high-quality text and images. This upgrade enhances creative capabilities for developers and content creators using Tencent's platform.
Teaching models to forget: Selective unlearning with Amazon Nova
AWS just introduced Nova, a new framework for selective unlearning in AI models. This allows developers to remove specific data from models without retraining them entirely, improving data privacy and compliance.
Tencent releases Hy3 open-source model that allegedly matches models up to five times its active size
Tencent just released the Hy3 open-source model, claiming it matches the performance of models five times its size. This could make advanced AI capabilities more accessible for developers and researchers.
GPT-4's dominance lasted a year while today's top models barely survive seven weeks at the top
GPT-4's reign as the leading AI model ended after just a year, with newer models struggling to maintain dominance for even seven weeks. This shift indicates a rapidly evolving landscape in AI development, where staying on top is becoming increasingly challenging.
What is Mistral AI? Everything to know about the OpenAI competitor
Mistral AI just launched as a competitor to OpenAI, focusing on developing advanced AI models. This move increases competition in the AI landscape, pushing innovation and potentially lowering costs for users.
Mistral's open-source Leanstral 1.5 aces formal math benchmarks and catches real bugs in code
Mistral just released Leanstral 1.5, which excels in formal math benchmarks and identifies real bugs in code. This update enhances the model's reliability for developers working on complex mathematical tasks and debugging.
Anthropic says it cut 80 percent of Claude Code's system prompt because Fable 5 models "want a smaller system prompt"
Anthropic just cut 80 percent of Claude Code's system prompt to align with Fable 5 models' preference for smaller prompts. This change aims to improve the model's performance and usability.
HippoRAG: Neurobiologically inspired RAG using Amazon Bedrock, Amazon Neptune, and personalized PageRank
AWS just introduced HippoRAG, a neurobiologically inspired retrieval-augmented generation model using Amazon Bedrock and Neptune. This new model enhances personalized search and information retrieval capabilities for users.