Mistral's Leanstral 1.5 Excels in Formal Math and Code Verification
Mistral AI's open-source Leanstral 1.5 model achieved 100% on miniF2F and solved 587 of 672 problems on PutnamBench.
Model Release
459 articles
New AI models as they ship — large language models, multimodal systems, and open-weight releases. Capabilities, context windows, pricing, and availability, summarized so you can compare what is actually new.
LATEST NEWS
-
-
Anthropic Cuts 80% of Claude Code System Prompt for Fable 5 Models
Anthropic reduced Claude Code's system prompt by 80% to align with Fable 5 models' preference for smaller prompts, according to a company insider.
-
Fable 5 AI Agents Now Complete 16 Percent of Freelance Jobs at Pro Quality
AI agents can now complete 16 percent of freelance jobs at professional quality, up from 2.5 percent eight months ago, according to the Remote Labor Index.
-
US lifts export bans on Anthropic’s Fable 5 and Mythos 5 models
The US has removed export restrictions on Anthropic’s Fable 5 and Mythos 5 models, following safety tests that cleared the AI systems for global deployment.
-
AWS GovCloud Now Supports NVIDIA Nemotron and OpenAI GPT OSS Models
AWS GovCloud (US) now supports NVIDIA Nemotron and OpenAI GPT OSS models, offering secure, compliant AI capabilities for U.S. government agencies.
-
Hugging Face Launches Pulpie, Pareto-Optimal Web Content Extractor
Hugging Face introduces Pulpie, a family of models achieving SOTA extraction quality at one twentieth the cost of Dripper, with pulpie-orange-small scoring 0.862 ROUGE-5 F1 on WebMainBench.
-
Google AI Unveils Gemini 3.5 Live Translate and New Hardware
Google released Gemini 3.5 Live Translate, supporting over 70 languages, in June 2026, enhancing real-time communication across devices.
-
Meta's Brain2Qwerty v2 Closes Gap With Surgical Implants
Meta's non-invasive brain-to-text AI, Brain2Qwerty v2, achieved a 39% average word error rate in a study with nine volunteers, outperforming its predecessor.
-
Google Launches Gemini Spark on Mac
Google has expanded Gemini Spark to macOS, adding real-time updates and integration with Google Tasks and Keep.
-
OpenAI Paper Details Three GPT-5.6 Pro Models
An OpenAI paper lists three GPT-5.6 Pro models for the first time, including Luna Pro, Terra Pro, and Sol Pro, marking a shift from a single top-tier strategy.
-
Claude Sonnet 5 Pricing Rises Behind Unchanged Token Rates
Claude Sonnet 5 ranks fifth in performance but costs more per task than Opus 4.8, despite unchanged token rates.
-
AWS Announces Bedrock Availability of Anthropic Claude Fable 5 Models
AWS announced today that Anthropic’s Claude Fable 5 models will be available on Bedrock, featuring enhanced guardrails to prevent misuse.