Revision history
What changed in the gateway, the catalog, and the rate card. Newest first.
SEC.01 — SHEET DATA
FIG.01 — SHEET TITLE BLOCK
Drawing no. — IDC-2026-07
Sheet — Changelog · 1 of 1 · NTS
Current rev — C.12 · issued
Baseline — REV A.1 ·
Entries — 16 across issues A, B, C
Scope — Gateway, model catalog, rate card
SEC.02 — REVISION BLOCK
Dates are the day a change landed in production, not the day upstream announced it. Retired IDs keep resolving through the alias window named in their entry.
FIG.02 — REVISION TREE
Moonshot's kimi-k3 is routable at 1M context with reasoning enabled on every call, so no thinking flag is required. Vision input is supported. The ID resolves to Moonshot's own endpoint, not a managed pool.
Upstream retired deepseek-chatdeepseek-reasonerdeepseek-v4-flash, reasoner traffic on deepseek-v4-pro. After that date the retired IDs return 404.
grok-4.5 was routable the day xAI shipped it, at 500K context with vision input. Catalog entry, rate card, and routing weights landed in the same deploy.
gemini-3.6-flash added and promoted to the default Google Flash route, displacing gemini-3.5-flash. The 3.5 ID stays in the catalog and keeps resolving; only the default target moved.
All three tiers — gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna — went live on OpenAI's launch day at 1M context. No 5.6 Codex variant shipped upstream, so gpt-5.3-codex remains the only code-tuned OpenAI ID in the catalog.
Zhipu glm-5.2 and MiniMax MiniMax-M3 added, both at 1M context. GLM-5.2 runs with reasoning on by default; MiniMax M3 takes vision input.
claude-fable-5 was routable on Anthropic's GA day. claude-opus-5 and claude-sonnet-5 stay in the catalog at their existing rates — Fable 5 is an addition to the generation, not a replacement for it.
qwen3.7-max and qwen3.7-plus added, both at 1M context with vision input. qwen3-max
Cohere command-a-plus-05-2026 added at 128K context. It is priced by contract rather than list, so the catalog shows Contact and the rate is fixed per agreement before the key is issued.
xAI retired grok-code-fast-1grok-build-0.1, which carries the same code-tuned role at 256K context, so existing integrations keep working without a string change.
deepseek-v4-pro and deepseek-v4-flash added, both at 1M context with reasoning on by default. These are the successors the C.11 alias window points at.
gpt-5.5-pro and mistral-medium-3-5 added. GPT-5.5 Pro is the most expensive ID in the catalog and the only one in its own rate band; Mistral Medium 3.5 lands at 256K context with vision input.
Per-call pricing now returns split across X-IDC-Rate-In and X-IDC-Rate-Out. Cost attribution no longer needs a lookup against your contracted rates at request time.
gemini-3.1-pro-preview opened as a preview channel at 1M context. Preview IDs have no stable upstream release behind them, so their rate and availability can change without a revision entry.
Qwen3.5 open weights brought up on idclinktech managed pools across three capacity tiers. Hosted rates include the serving margin and are quoted independently of Qwen's own API IDs.
Sheet baseline, kept for the record. The catalog opened on the Claude 4.7 / 4.6 and GPT-5.4 / 5.5 generations; every ID issued at REV A has since been superseded or aliased forward, and none are still listed.
SEC.03 — NOTES ON THIS SHEET
Rev letter = sheet issue · rev number = entry within that issue
Struck IDs are retired upstream · the successor ID is named in the same entry
Legacy = still resolvable, no longer listed, rate frozen at its last contracted value
Preview channels may change rate or availability without a revision entry