Published: · Last updated:
At Chat-O, we build the ultimate AI Chat and Productivity tool, constantly enhancing your experience with fresh features and cutting-edge models. This May 2025 update laid the foundation for everything that followed—including Grok 3, Gemini 2.5, o3, and o4, GPT-5, and the latest Kimi K3.
We welcomed several powerful additions:
Google’s most capable model at the time, featuring a 1M token context window and native multimodal understanding. It could process entire codebases, long documents, and hours of video in a single request. This was the precursor to later Gemini iterations.
A specialized model for mathematical theorem proving and formal verification. DeepSeek Prover V2 excelled at complex proofs in Lean and Isabelle, making it invaluable for researchers and mathematicians. Its formal reasoning capabilities complemented general-purpose models in our lineup.
Inception Labs’ fast, efficient coding model optimized for real-time code generation. Mercury Coder Small Beta prioritized low latency while maintaining solid code quality, ideal for interactive coding sessions and IDE integration.
Mistral AI’s balanced mid-size model offering strong performance across languages, particularly European languages. At roughly 70B parameters, it provided an excellent quality-to-cost ratio for multilingual applications.
Microsoft’s compact 14B model with enhanced reasoning capabilities. Despite its small size, Phi 4 Reasoning Plus competed with models 5-10x larger on reasoning benchmarks, demonstrating the potential of specialized small models—a philosophy later seen in GLM-4.7. It was particularly effective for structured reasoning tasks like code debugging, logical puzzles, and multi-step math problems where larger models often over-engineer solutions.
| Model | Parameters | Best For | Context |
|---|---|---|---|
| Gemini 2.5 Pro | Unknown | Multimodal, long context | 1M tokens |
| DeepSeek Prover V2 | Unknown | Math proofs | 32K tokens |
| Mercury Coder Small | ~7B | Fast code gen | 16K tokens |
| Mistral Medium 3 | ~70B | Multilingual | 32K tokens |
| Phi 4 Reasoning Plus | 14B | Compact reasoning | 16K tokens |
Our team integrates new models swiftly, often the same day they’re announced, and we refine our selection based on your feedback. This commitment has brought you GPT-OSS-120B, the Complete Kimi Family, and more.
We overhauled our Settings page with a powerful feature: enable or disable specific models for each tier. While our multi-model approach—where AIs collaborate for better answers—offers superior results, we champion your right to fine-tune your experience.
This customization layer has become even more valuable as our lineup expanded. Today you can choose from the GLM-5.2 for million-token reasoning, Kimi K3 for extreme parameter counts, or compare them side by side in our GLM-5.2 vs Kimi K3 comparison.
Your privacy is paramount. We do not access your message content, and we don’t use providers that access messages or use prompts to improve their models. Model performance analysis relies on anonymized statistics only. This commitment is detailed in Why We Don’t Compromise on Privacy and backed by our content reporting system.
This May 2025 update was a pivotal moment. Since then, we’ve added:
The model customization feature has proven especially popular with power users. Developers working across multiple domains create custom presets—for example, routing code questions to Mercury Coder, document analysis to Gemini 2.5 Pro, and mathematical proofs to DeepSeek Prover V2—all within a single session. This flexibility, combined with our privacy-first approach, makes Chat-O a uniquely versatile AI workspace.
The ability to toggle models per tier has become even more critical as our lineup grew from 5 models to over 20. Users can now curate their personal model selection from the Complete Kimi Family and Complete Z-AI GLM Family, keeping only the models they actually use.
We’re excited for you to explore these models and take greater control of your Chat-O experience. The journey from May 2025 to today shows how fast AI evolves—and we’ll keep bringing you every breakthrough.