#2570 docs: MODELS.md's Coder reads long prompts fast, not "the fastest of all" (Q2_0 reads faster on the RTX 5070)

本文

**What it changes.** One sentence in `docs/MODELS.md`'s Coder paragraph. It said the Coder "reads long prompts the fastest of all". The page's own speed tables (32K prompts) say:
- RTX 5070: Q2_0 2,650 tokens/s, Coder 2,180, IQ2_XS 2,090 (Q2_0's row is engine 0.1.36, the others 0.1.26; the claim predates Q2_0's 0.1.36 numbers);
- RX 9070 XT: Coder 1,420, Q2_0 1,160, IQ2_XS 1,110.

It now reads "second only to Q2_0 on the RTX 5070 above, the fastest of the three on the RX 9070 XT".

**What it leaves alone.** Every number and every other sentence (a two-line diff plus the rewrapped line after it).

**How it was checked.** Against the two tables on the same page; a sub-agent reviewer checked the wording against them.

🤖 Generated with [Claude Code](https://claude.com/claude-code)