GAP-L1-P0-claude-models-pricing-and-lineup

Current state

• schema/examples/compute/models/claude-opus-4-7.yaml has costPer1kInputTokens: 15.0 and costPer1kOutputTokens: 75.0 with TODO. Real pricing is **\$5 / \$25 per MTok** (i.e. 0.005 / 0.025 per 1k).
• Attribute name says "per1k" but the schema's attributeType is cost-per-million-tokens — naming mismatch.
• No example for Claude Sonnet 4.6 (claude-sonnet-4-6, $3/$15, 1M ctx, released ~Q1 2026 with extended thinking).
• No example for Claude Haiku 4.5 (claude-haiku-4-5-20251001, $1/$5, 200k ctx).
• No example for Opus 4.6, Opus 4.5, Sonnet 4.5, Opus 4.1.
• releaseDate: "2026-01-15" for Opus 4.7 is a placeholder.

Desired state

• Correct Opus 4.7 pricing: costPer1kInputTokens: 0.005, costPer1kOutputTokens: 0.025 (or rename attribute to costPerMTokInput/Output to match attribute-type name).
• Add 5 ModelVersion examples: claude-sonnet-4-6, claude-haiku-4-5, claude-opus-4-6, claude-opus-4-5, claude-sonnet-4-5 (legacy still served).
• Add adaptiveThinking: bool attribute on ModelVersion (Opus 4.7 + Sonnet 4.6 = true; Haiku 4.5 = false). Distinguish from supportsThinking (extended thinking).
• Add tokenizerVersion attribute (Opus 4.7 uses a new tokenizer per Anthropic docs — affects context window vs char count).
• Add aliasIds: list<string> (e.g. claude-opus-4-7 alias vs. dated snapshot).
• Add bedrockId, vertexId as first-class attributes (currently only in regions).

Evidence

• https://platform.claude.com/docs/en/docs/about-claude/models/overview (confirms Opus 4.7 = \$5/\$25, 1M ctx, no extended thinking but adaptive thinking, training cutoff Jan 2026)
• https://platform.claude.com/docs/en/about-claude/model-deprecations (Sonnet 4 / Opus 4 retire 2026-06-15)

Propagation status

• Level 1 (real-world vs graph): open
• Level 2 (graph vs docs): not-started — add adaptiveThinking invariant row to 02-node-kinds/compute-path.md and to coverage-checklist Layer 1 row
• Level 3+: cascade

Propagation chain

• Level 1: fix pricing typo, add new example files, add 2 attributes (adaptiveThinking, tokenizerVersion).
• Level 2: rename costPer1k* → costPerMTok* OR change attributeType reference; either way the markdown spec disagrees with the YAML.

Notes

The pricing typo by itself makes derived cost calculations wrong by 3000x. P0.