DeepSeek V4 Teased as Qwen 3.5 Matches Top-Tier Models in Reasoning Benchmarks
New AI Models & Releases
DeepSeek V4 Release Imminent: DeepSeek has announced the upcoming release of V4, hinting at significant advancements in performance and groundbreaking innovation. The announcement has sparked speculation regarding the model's potential to disrupt the current market once it becomes available.
Cohere Releases Tiny Aya Multilingual Model: Cohere has introduced "Tiny Aya," an open-weights 3.35 billion parameter model optimized for efficient and balanced multilingual representation across over 70 languages. This research release aims to provide strong performance in a compact size, with weights available in several versions on Hugging Face.
- Tiny Aya
- https://cohere.com
- https://cohere.com/research
- https://cohere.com/cohere-labs-cc-by-nc-license
- https://docs.cohere.com/docs/c4ai-acceptable-use-policy
- https://cohere.com/blog/cohere-labs-tiny-aya
- https://github.com/Cohere-Labs/tiny-aya-tech-report/blob/main/tiny_aya_tech_report.pdf
- https://huggingface.co/CohereLabs/tiny-aya-earth-GGUF
- https://huggingface.co/CohereLabs/tiny-aya-fire-GGUF
- https://huggingface.co/CohereLabs/tiny-aya-water-GGUF
- https://huggingface.co/CohereLabs/tiny-aya-global-GGUF
Benchmark Results & Model Analysis
Qwen 3.5 Performance Evaluated Across Multiple Benchmarks: Recent analysis of Qwen 3.5 reveals mixed results, with the model failing financial simulations on Vending-Bench 2 while showing major improvements in spatial reasoning on MineBench. In reasoning tasks, Qwen 3.5 reportedly performs on par with top-tier models like Claude Opus 4.6 and GPT-5.2.
- Qwen 3.5 goes bankrupt on Vending-Bench 2
- Difference Between QWEN 3 Max-Thinking and QWEN 3.5 on a Spatial Reasoning Benchmark (MineBench)
Industry News & Talent
OpenClaw Creator Hired by OpenAI: Peter Steinberger, the creator of the OpenClaw project, has joined OpenAI only 90 days after launching his initiative. This hiring move suggests that features from OpenClaw may eventually be integrated into OpenAI’s ChatGPT Codex service.