Replace the GPU price-based cost in the GLM-5.3 post with output per GPU-hour - #4486

Follow-up to #4483. Partners asked us not to use SemiAnalysis's $4.00 per chip-hour, so the post no longer assumes any GPU price.

• The paragraph under the intro now says each GPU-hour of the fleet produces 313k output tokens, which would cost $13.51–20.71 to buy from the three endpoints. Readers compare that with their own GPU-hour cost, whether rented or owned.
• The cost chart tab is removed, and the note explains the per-GPU-hour figure instead of the $4.00 rate.
• Limitation 3 now says the comparison leaves out what a GPU-hour costs.

Numbers: 4,872.19 output tok/s × 3,600 ÷ 56 GPUs = 313.2k output tokens per GPU-hour; × $43.13 and $66.13 per 1M = $13.51 and $20.71.

Stacked on #4478.

🤖 Generated with Claude Code