Lab Reports and AIHackers.net Living Notes

Field notes for 2026-02-04. Stack snapshot, pricing, and what I’m testing next.

Present Stack

  1. Tools: VSCode, Claude Code, Codex
  2. Models: Opus 4.5, Kimi k2.5 (free via OpenCode Zen or near-free via trial paths), GPT-5.2-codex, GLM 4.7 when limited, Gemini 3 Flash
  3. Research stack: Obsidian + Deep Research with GPT and Kimi for specs, comparisons, and recording data

System Architecture

  1. Agent work runs on a Debian VM with UFW + Tailscale. Isolation-first mindset, disposable VM/VPS when needed (see YOLO Safely). Tailscale writeup coming.

Plans

Now

  1. Demonstrate early “yolo” experiment leading to MCP exploit (see MCP First Contact field notes - draft).
  2. Document the full “YOLO Safely” deployment from scratch — $5/mo VPS is safer isolation than a $600+ dedicated box. I am not buying a machine just to run OpenClaw. Disposable VPS is the better strategy. VPS playbook.
  3. Demonstrate Codex’s new app (tools/codex).

Next

  1. Review and re-create every article.
  2. Start building foundations for a community. Might be lonely, but posting regularly is good practice and maybe someone wants to chat.
  3. The best way to learn is by teaching + doing. Go do crazy stuff, then distill it into actionable, entertaining posts.
  1. Claude $20: Claude Code + Opus 4.5. $200 max plan would be second choice, skipping for cost reasons.
  2. ChatGPT $20: More generous limits than Opus 4.5 with GPT‑5.2‑codex. I rate it third best (Opus 4.5, Kimi k2.5, GPT‑5.2‑codex).
  3. Kimi $3.49/mo: K2.5 via haggling/trial paths. Details: /value/kimi-access/.

Pricing strategy refs: /value/free-stack/, /value/smart-spend/

Shower Thoughts

That is it. Along with free offers, it has been nearly enough. Doing research + trying promos gave more compute than I knew what to do with in the first few days, but the landscape changes almost daily and the limits are tightening.

OpenAI releasing Codex felt like an answer to the Kimi ecosystem — deep research, slide creation, document creation, and strong frontier performance. This puts real pressure on incumbents chasing massive funding targets. The model economics war has begun; I’m watching it closely.

Recent Changes (Feb 2026)

  • 2026‑02‑03: OpenCode Zen limits on Kimi k2.5 + GLM 4.7 seem dramatically reduced. Generous limits → practically none.
  • 2026‑02‑03: Kilo Code’s free Kimi k2.5 option appears to be gone (at least on my side). Tracking: /value/kimi-access/.
  • 2026‑02‑01: Kiro CLI removed Opus 4.5 from free use. Now Sonnet 4.5 only, 500 credits month one, 50 credits/mo after. Still useful free top‑tier, but worse value. Tool page.
  • 2026‑02‑04: Leaning more toward AGENTS.md vs skills. Skills adherence is low; AGENTS.md gets respected. Considering multiple AGENTS.md per project with progressive disclosure. Documented results in practical guide.

Notes

Kimi pricing haggling details (including $0.99 first month) live here: /value/kimi-access/.