fix(ai): omit the default service tier, reprice cache writes from message_delta, repoint the zai default

[snimu]

Summary

Three provider wire/pricing defects:

• service_tier: "default" is no longer sent on the wire (OpenAI Responses and Codex Responses). For OpenAI, absence is the default, so this changes nothing there — but strict endpoints such as GitHub Copilot reject the explicit field, failing the whole request. Fixes the defect reported in discussion [Bug] Error: Provider rejected the request (invalid_request_error, 400): service_tier is not supported #1413.
• Anthropic cache writes are repriced from message_delta. The cache-write rate was computed once (at message_start, or before the stream), and message_delta billed its updated cache-write token count at that stale rate. When a delta carries a cache_creation breakdown (5m/1h mix), the rate is now recomputed exactly as message_start does; without a breakdown the prior rate is kept. Fixes the defect reported in discussion [Bug] Anthropic reports stale cache-write cost when message_delta updates cache creation #1769.
• The zai default model exists again. defaultModelPerProvider.zai pointed at glm-5.1, which was removed from the catalog — every default zai selection silently degraded to a fallback template model. The default is now glm-5.3, and a new catalog-existence test makes any future default/catalog drift a CI failure instead of a silent downgrade. Fixes the defect reported in discussion [Bug] Stale default model for zai provider (glm-5.1 removed from catalog) silently falls back to outdated glm-4.7 #1679.

Validation

• four pins, each verified fail-unfixed: default-tier omission on the OpenAI Responses body and on the Codex body (one per wire builder), message_delta breakdown repricing (exact-cost assertion), and per-provider default catalog existence
• suites: anthropic-sse-parsing (7), openai-responses-copilot-provider (24), openai-codex-stream (18), model-resolver (25) — all green, sanitized env
• root npm run check passes via the pre-commit hook

LOC

Src +20/−3 across four files (all point fixes; the only added mechanism is the 8-line delta reprice mirroring the existing message_start path). Tests +143/−2, changelog 2 fragments.

Linear: RES-1266 https://linear.app/primeintellect/issue/RES-1266

---

Note

Low Risk
Targeted provider request shaping and usage-cost fixes plus a default model ID update; no auth or broad API redesign.

Overview
Fixes three wire/pricing/defaults issues across pi-ai and coding-agent.

OpenAI Responses and Codex no longer put service_tier on the request body when the tier is "default". Omitting the field is the real default; sending it explicitly breaks strict proxies (e.g. GitHub Copilot). Non-default tiers (flex, priority) are unchanged.

Anthropic streaming now recomputes cache-write cost on message_delta when usage includes a cache_creation breakdown (5m vs 1h), matching the message_start path. Final token counts from the delta no longer get billed at a stale rate from the start event.

zai default model moves from glm-5.1 (removed from the catalog) to glm-5.3. A test asserts every defaultModelPerProvider entry exists in the catalog so future drift fails in CI instead of silently falling back to a template model.

^{Reviewed by Cursor Bugbot for commit 64e5f3d. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Omit default service_tier on OpenAI requests, reprice Anthropic cache writes from message_delta, point zai default to glm-5.3

• Omits service_tier from OpenAI Responses and Codex request bodies when the tier is "default"; explicitly requested non-default tiers are still serialized.
• Anthropic streaming now reads the cache-creation breakdown from message_delta usage and recalculates cache-write cost using the configured five-minute or one-hour retention tier, replacing the earlier message_start-derived price.
• Changes the zai provider default model from glm-5.1 to glm-5.3 in model-resolver.ts.
• Risk: Anthropic cache-write cost now depends on message_delta containing the cache-creation breakdown; if a response omits it, cache-write cost may be zero or stale. zai consumers relying on glm-5.1 as the default will silently move to glm-5.3.

^{Macroscope summarized 64e5f3d.}