# This week in LLM prices (2026-10-21 to 2026-10-28)

104 models added, 24 repriced, 11 got a retirement date, 0 removed, in LiteLLM's catalogue of 4,344 listed models.

**Retirement dates set this week, soonest first**
- together_ai/MiniMaxAI/MiniMax-M3: 2026-11-06
- together_ai/thinkingmachines/Inkling: 2026-11-06
- github_copilot/claude-opus-4.8-fast: 2026-11-16
- azure_ai/Phi-4: 2027-01-17
- azure_ai/Phi-4-mini-instruct: 2027-01-17

**New this week, already ranked on Arena**
- google / gemini-4-argon: #1 for overall
- moonshot / kimi-k3.1-pro: #8 for webdev
- openai / gpt-6.1-terra: #14 for webdev
- moonshot / kimi-k3: #18 for overall
- tencent / hy4-preview: #22 for webdev

Top quality costs 34.4x less than a year ago: a model as good as last year's #1 (Gemini 2.5 Pro, then $3.44 per 1M tokens) is GLM-5.3 (Max) at $0.10.

Price changes that look like corrections of the catalogue (a model's first week, from zero, or 10x either way) are left out of the cuts and rises. Not financial or procurement advice.
Every model, every change, the method: https://pouriaetab.github.io/llmledger/