Update model costs

[anchpop]

Summary

Checked the model pricing table in src/model_prices.rs against current provider documentation.

• Anthropic (Claude): no changes — all listed prices match current published rates.

• OpenAI: added the new GPT-5.6 family (Sol, Terra, Luna), released July 9, 2026:

  • gpt-5.6 (bare alias, routes to Sol): $5.00 input / $0.50 cached / $30.00 output per 1M tokens
  • gpt-5.6-sol: $5.00 / $0.50 cached / $30.00 (same price point as gpt-5.5)
  • gpt-5.6-terra: $2.50 / $0.25 cached / $15.00 (same price point as gpt-5.4)
  • gpt-5.6-luna: $1.00 / $0.10 cached / $6.00 (new lower-cost tier)

  All other OpenAI prices match current published rates.

• Google Gemini: four Gemini 3.x models were missing cached_input pricing (previously None). Added them at the standard ~10% of input-token rate (matching the pattern already used for the other Gemini entries in this file):

  • gemini-3.1-pro: cached input $0.20/1M tokens
  • gemini-3-flash: cached input $0.05/1M tokens
  • gemini-3.1-flash-lite: cached input $0.025/1M tokens
  • gemini-3.5-flash: cached input $0.15/1M tokens

Sourcing note

Direct fetches of platform.openai.com/docs/pricing, openai.com, and ai.google.dev/gemini-api/docs/pricing were blocked by this environment's network policy, so pricing was cross-checked via search and secondary sources (multiple independent trackers/aggregators that agreed on the same GPT-5.6 figures, plus Google Cloud's Gemini Enterprise Agent Platform pricing page for the Gemini cached-input rates). The GPT-5.6 pricing/model-ID pattern was also confirmed by the user directly.

A few things surfaced during research that are not included in this PR because they couldn't be confirmed with high confidence:

• A possible discrepancy on gpt-5.2-pro pricing — the file's existing $21/$168 fits the established multiplier pattern from gpt-5-pro better than an unconfirmed $10.50/$84 figure found in one low-quality source, so it was left unchanged.
• Several already-deprecated/retired OpenAI models (gpt-4o-audio-preview, gpt-4o-realtime-preview, gpt-4o-mini-audio-preview, gpt-4o-mini-realtime-preview, gpt-4.5-preview) remain in the table for historical cost calculations; pricing for these is unchanged since they're no longer callable.

Test plan

• cargo test model_prices passes (4/4 tests)
• Reviewed diff for scope
• Verified the longest-prefix-match lookup still resolves gpt-5.6, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna to their correct distinct rows

Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com