A few notes and links for large language models (LLMs) as of Feb 2026.

Use Cases

Divergent Thinking

For divergent thinking, I use Anthropic’s Claude Opus 4.5 (via Claude Code).

I like its agency: it takes initiative and completes tasks without much back-and-forth unless you use Plan Mode.

• Claude Opus 4.5 system card (PDF)

Converging Thoughts → Writing Code

Once I have a plan, I switch to OpenAI’s GPT-5.2-Codex (via the Codex CLI).

• GPT-5.2-Codex system card addendum (PDF)

Rough workflow:

1. Claude Opus 4.5 for initiative and divergence (get something moving)
2. GPT-5.2 Codex for implementation (write the code)

2026-Q1 Gen-AI Guide (Personal Workflow)

Vendor	Models (January 2026)	Max Context	Best At
OpenAI	GPT-5.2 (Codex) (system card (PDF))	—	Code writing & implementation
	GPT-4.1	1M tokens	Code & structured work, logic
	o3	200k	Creative writing, brainstorming, ideation
Google	Gemini 2.5 Pro	1M	Code, large codebases, large context prompts
	Gemini 2.5 Flash (docs)	128k	Fast responses
Anthropic	Claude 4 Sonnet	200k	Code & structured work, logic
	Claude Opus 4.5 (system card (PDF))	—	Agentic workflows, ideation, divergence
	Claude 3.7 Sonnet	200k	Code & structured work, logic
Meta	Llama 4 Scout (open)	10M/1M	Open Weight LLM
xAI	Grok 3	130k	Reasoning, code & structured work, logic
	Grok 4	256k	Reasoning, coding, multimodal
DeepSeek	V3-0324, R1-0528	64k	Cost-sensitive experimentation

OpenAI

• GPT-5.2-Codex - My default for implementation (via Codex CLI)
  • System card addendum (PDF)

Anthropic

• Claude (Wikipedia)
• System cards (PDF)
• Claude Opus 4.5 - My default for agentic/divergent work (via Claude Code)
  • System card (PDF)
• Claude 4.5 Sonnet - Sept 29, 2025

Previous Anthropic models

- **Claude 4 Sonnet** - July 3, 2025 - **Claude 3.7 Sonnet** - February 24, 2025 ([system card](https://www.anthropic.com/claude-3-7-sonnet-system-card)) - **Claude 3.5 Sonnet** - June 20, 2024

Google

• Gemini (Wikipedia)
• Gemini 3 Pro - Nov 18, 2025
• Gemini 3 Pro Image - Nov 20, 2025

Previous Google models

- **Gemini 2.5 Pro** - June 2025 ([model card](https://storage.googleapis.com/model-cards/documents/gemini-2.5-pro-preview.pdf))

• Google DeepMind Model Cards

Decoding OpenAI’s Labels

Label	Translation
o*	“Omni” = multimodal & agent-ready
4.1	Latest GPT-4 weights
mini / nano	Fewer parameters → lower cost & latency

Snapshot suffixes (e.g., -0425) indicate frozen weights date.

Apps and Links

• AI Studio - Google’s Gemini playground
• NotebookLM - Google’s tool for working with PDFs, text, YouTube videos
• OpenAI Playground
• ChatGPT
• Claude
• Meta Llama
• xAI Grok

Notes

1. Open-weight models (June 2025 - Feb 2026):
   • Kimi K2.5 - Moonshot AI
   • Kimi K2 Thinking - Moonshot AI
   • Qwen 3 - Alibaba Qwen
     • Qwen Chat
   • DeepSeek R1/V3.2 - DeepSeek
   • GLM-4.7 - Zhipu AI
   • MiniMax M2.1 - MiniMax
   • Llama 4 Scout - Meta
   • Mistral Small 3 - Mistral AI
2. synthetic.new - API provider for open-weight models (GLM-4.7, MiniMax M2.1).
   • https://synthetic.new/policies/privacy

---

• “Context” refers to how many tokens the model can process in a single request. For Anthropic models, see their context windows documentation.