Prices are per 1 million tokens in USD, standard (non-batch) tier, verified 2026-09-04 against official pricing pages. Claude Fable 5.1: Released Sep 1, 2026. Anthropic's Mythos-class flagship tier above Opus 5 — same $10/$50 base rate as Fable 5, but cache reads drop to 2.5% of the input price ($0.25/MTok vs the 10% every other Claude model charges). Shares its underlying model with Claude Mythos 5.1, offered only to approved organizations. Claude Fable 5: Superseded by Claude Fable 5.1 (Sep 1, 2026), which keeps the same base rate but cuts cache reads to $0.25/MTok. Fable 5 remains available as a pinned snapshot with cache reads at the standard 10% ($1.00/MTok). Shares its underlying model with Claude Mythos 5. Claude Sonnet 5: The $2 / $10 launch price is now permanent; the planned Sep 2026 increase to $3 / $15 was cancelled. GPT-5.6 sol: Promotional price, available at least through Nov 21, 2026 (previously $5 / $30). Requests beyond the long-context threshold bill at 2× input / 1.5× output. GPT-5.6 terra: Requests beyond the long-context threshold bill at 2× input / 1.5× output. GPT-5.6 luna: Requests beyond the long-context threshold bill at 2× input / 1.5× output. Gemini 3.1 Pro Preview: Prompts over 200K tokens bill at $4 / $18. Gemini 3.6 Flash: Intro price through Dec 31, 2026; $1.50 / $7.50 from Jan 1, 2027. Grok 4.6: Requests over 200K tokens bill at 2×. Grok 4.5: Requests over 200K tokens bill at 2×. Kimi K3: Always-on reasoning; output includes thinking tokens. DeepSeek V4 Pro: Peak-hour price (Beijing weekdays 09:00–12:00, 14:00–18:00) effective Aug 17, 2026; all other hours bill at 50% of peak. DeepSeek V4 Flash: Peak-hour price (Beijing weekdays 09:00–12:00, 14:00–18:00) effective Aug 17, 2026; all other hours bill at 50% of peak.

Batch discount is applied only to models whose provider offers an async batch tier. Cached-input pricing models the read price; cache-write surcharges (Anthropic) are not included. Long-context surcharges apply above provider thresholds and are noted per model.