wednesday, july 22, 2026
top 5 trending●previous 24h●generated 14d ago
Claude Fable 5 assisted in discovering a counterexample to the Jacobian Conjecture, a long-standing open problem in mathematics. The model generated mathematical reasoning that led to disproving the conjecture, with the work also verified through conversation with mathematician Terence Tao.
- Claude Fable 5 produced a counterexample disproving the Jacobian Conjecture
- Terence Tao engaged in ChatGPT conversation discussing the counterexample
- Codex-based reproducible proof documented against the Jacobian Conjecture
- Claude Fable stylistically closer to Kimi K3 than Claude Opus
- Prompt crafting increasingly over-rated; direct requests more effective
OpenAI released GPT-5.6 with multiple frontier model variants (Sol Pro, Sol, Terra, Luna) designed for different use cases. Early testing shows GPT-5.6 Sol Pro outperforming competitors on creative and mathematical reasoning tasks, including discovering a false mathematical conjecture with minimal prompting.
- GPT-5.6 Sol Pro variant demonstrated superior performance on creative writing tasks compared to Fable, Kimi, and Gemini
- GPT-5.6 Pro discovered the Dinitz-Garg-Goemans conjecture is false using only 58 words of prompting
- Multiple GPT-5.6 variants available: Sol Pro, Sol, Terra, and Luna for different applications
- V0.5.2 release included 1400 fixes addressing SOA, ownership, and garbage collection improvements
Kimi K3, a Chinese AI model from Moonshot AI, has achieved state-of-the-art performance competitive with Fable, reportedly solving the Jacobean Conjecture and demonstrating advanced reasoning capabilities with extended chain-of-thought outputs. The breakthrough has elevated Moonshot AI's valuation to $50 billion ahead of a planned Hong Kong IPO, marking a significant shift in the competitive balance of the global AI ecosystem toward Chinese labs.
- Kimi K3 matches or exceeds Fable's performance on benchmarks, now considered state-of-the-art
- Fable reportedly proved the Jacobean Conjecture false, a major mathematical achievement
- Extended reasoning: K3 generated 32-page chain-of-thought responses on complex creative tasks
- Moonshot AI targeting $50 billion valuation with Hong Kong IPO following K3 release
- Chinese AI labs' approach and capabilities now recognized as competitive alternative to San Francisco-centered development
Google released Gemini 3.6 Flash alongside updated variants 3.5 Flash-Lite and 3.5 Flash Cyber. The release deprecated temperature, top_p, and top_k parameters across Gemini models.
- Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber released
- Temperature, top_p, and top_k parameters deprecated and ignored in Gemini models
- 3.6 Flash tested on shader benchmark with performance comparable to prior versions