Fix model cost units: dollars per million tokens, not per token

[CNQQC]

Problem

src/models.ts declares model costs as dollars per token (input: 1e-6 for Haiku), but opencode and models.dev express cost.* as dollars per million tokens — opencode divides by 1e6 itself when it multiplies a cost by a token count.

models.dev's own entry for the same model:

anthropic/claude-haiku-4-5 -> {"input": 1, "output": 5, "cache_read": 0.1, "cache_write": 1.25}

Net effect: every session cost opencode reports through this provider is exactly 1,000,000x too low, so the cost indicator sits at $0.00 no matter how much you run. Token counts — including the cache read/write split — were already correct; only the dollar amount was wrong.

This is easy to miss because subscription users aren't billed per token anyway. But the figure is still the only in-TUI signal for how fast a session is draining a Max/Pro plan, which is exactly what the (N×) multiplier in the display name is there to approximate.

Fix

Multiply the five cost constants by 1e6 so they read as the published per-million-token prices ($1/$5 Haiku, $3/$15 Sonnet, $2/$10 Sonnet 5, $5/$25 Opus, $10/$50 Fable/Mythos). Comment updated to name the unit and cite the models.dev entry so it doesn't regress.

The (N×) multiplier suffix is unaffected — it comes from the input/output price ratios, which don't change.

Verification

bun run typecheck clean, bun run build clean, bun run test — 294/294 pass (the two exact cost assertions in test-config-models.ts updated to the new unit).

End-to-end against opencode 1.18.12, real Haiku 4.5 turn, 10 input / 62 output / 10,583 cache write / 15,973 cache read:

	reported cost	reported ÷ actual
before	$0.00000002	0.000001
after	$0.01514605	1.000000

Note on #22 / #23

This touches sonnet5Cost, which #23 also rewrites for the September 1 standard-pricing bump. Trivial conflict either way — after this lands, that PR's target values become { input: 3, output: 15, cacheRead: 0.3, cacheWrite: 3.75 }.

🤖 Generated with Claude Code

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.