ZAI and Kimi provider endpoint routing fixes

NousResearch/hermes-agent · cluster 966 · agent

Data as of 2026-10-21 12:10 UTC

12 open items. Fix merged in #9907. Check it covers your case.

open issues

open PRs

items grouped

Known?: Yes. 31 related items grouped here, 8 issues already closed.
Fix event: A fix was merged to main in PR #9907. Check it covers your case. No release data is tracked.
Tracker: The maintainers' tracker marks this partially resolved, deciding PR #9907. That is separate from the fix event.
Still open: 4 issues and 8 PRs.

Next step: read #9907, then check whether your report is one of the 4 still open.

Read #9907 Pick up a Lab task

These issues and PRs address incorrect endpoint selection, ignored base URL configuration, and wrong billing paths for the ZAI/GLM and Kimi/Kimi-Coding providers across authentication, credential pools, and vision auxiliary clients. A recurring theme is hardcoded or incorrectly ordered endpoint probing causing users to be charged against metered quotas instead of subscription/coding-plan quotas.

Open issues4

Issue#5908kimi-coding credential pool base_url not re-resolved from key prefix on load
Issue#18302[Bug]: ZAI/GLM provider base_url causes 429/404 errors — config ignored by 5 code paths
Issue#42536[bug] Z.AI/GLM provider endpoint probe order causes Coding Plan users to be charged against wallet balance instead of subscription quota
Issue#55112[Bug]: hardcoded zai vision base_urls (paas/v4) bypass the coding-plan endpoint → silent metered billing

Open PRs8

PR#13993fix(providers): add moonshotai provider, disambiguate from kimi-for-coding
PR#18694fix(auth): re-resolve Kimi pool base URLs
PR#37816Respect configured ZAI base URL
PR#42547fix(zai): probe Coding Plan endpoints before generic to prevent silent wallet drain
PR#55116fix(auxiliary): prepend zai coding-plan vision endpoint before metered paas/v4 (fixes #55112)
PR#55156fix(auxiliary): try zai coding-plan vision endpoint before metered fallbacks (#55112)
PR#83537fix(auth): re-resolve stale kimi pool base_url on load (#5908)
PR#114111fix(auth): route Kimi Code credentials by key

Fixes on record

Merged to main. A stamp counts the issues a fix is linked to.

#9907 feat(zai): add GLM-5V-Turbo support for coding plan

merged to main

files: agent/auxiliary_client.py, hermes_cli/auth.py, hermes_cli/models.py

Show the diff

agent/auxiliary_client.py

@@ -112,6 +112,7 @@ def _normalize_aux_provider(provider: Optional[str]) -> str:
 # "exotic provider" branch checks this before falling back to the main model.
 _PROVIDER_VISION_MODELS: Dict[str, str] = {
     "xiaomi": "mimo-v2-omni",
+    "zai": "glm-5v-turbo",
 }
 
 # OpenRouter app attribution headers

hermes_cli/auth.py

@@ -383,13 +383,16 @@ def _resolve_api_key_provider_secret(
 # Z.AI has separate billing for general vs coding plans, and global vs China
 # endpoints.  A key that works on one may return "Insufficient balance" on
 # another.  We probe at setup time and store the working endpoint.
+# Each entry lists candidate models to try in order — newer coding plan accounts
+# may only have access to recent models (glm-5.1, glm-5v-turbo) while older
+# ones still use glm-4.7.
 
 ZAI_ENDPOINTS = [
-    # (id, base_url, default_model, label)
-    ("global",        "https://api.z.ai/api/paas/v4",        "glm-5",   "Global"),
-    ("cn",            "https://open.bigmodel.cn/api/paas/v4", "glm-5",   "China"),
-    ("coding-global", "https://api.z.ai/api/coding/paas/v4",  "glm-4.7", "Global (Coding Plan)"),
-    ("coding-cn",     "https://open.bigmodel.cn/api/coding/paas/v4", "glm-4.7", "China (Coding Plan)"),
+    # (id, base_url, probe_models, label)
+    ("global",        "https://api.z.ai/api/paas/v4",        ["glm-5"],   "Global"),
+    ("cn",            "https://open.bigmodel.cn/api/paas/v4", ["glm-5"],   "China"),
+    ("coding-global", "https://api.z.ai/api/coding/paas/v4",  ["glm-5.1", "glm-5v-turbo", "glm-4.7"], "Global (Coding Plan)"),
+    ("coding-cn",     "https://open.bigmodel.cn/api/coding/paas/v4", ["glm-5.1", "glm-5v-turbo", "glm-4.7"], "China (Coding Plan)"),
 ]
 
 
@@ -397,35 +400,37 @@ def detect_zai_endpoint(api_key: str, timeout: float = 8.0) -> Optional[Dict[str
     """Probe z.ai endpoints to find one that accepts this API key.
 
     Returns {"id": ..., "base_url": ..., "model": ..., "label": ...} for the
-    first working endpoint, or None if all fail.
+    first working endpoint, or None if all fail.  For endpoints with multiple
+    candidate models, tries each in order and returns the first that succeeds.
     """
-    for ep_id, base_url, model, label in ZAI_ENDPOINTS:
-        try:
-            resp = httpx.post(
-                f"{base_url}/chat/completions",
-                headers={
-                    "Authorization": f"Bearer {api_key}",
-                    "Content-Type": "application/json",
-                },
-                json={
-                    "model": model,
-                    "stream": False,
-                    "max_tokens": 1,
-                    "messages": [{"role": "user", "content": "ping"}],
-                },
-                timeout=timeout,
-            )
-            if resp.status_code == 200:
-                logger.debug("Z.AI endpoint probe: %s (%s) OK", ep_id, base_url)
-                return {
-                    "id": ep_id,
-                    "base_url": base_url,
-                    "model": model,
-                    "label": label,
-                }
-            logger.debug("Z.AI endpoint probe: %s returned %s", ep_id, resp.status_code)
-        except Exception as exc:
-            logger.debug("Z.AI endpoint probe: %s failed: %s", ep_id, exc)
+    for ep_id, base_url, probe_models, label in ZAI_ENDPOINTS:
+        for model in probe_models:
+            try:
+                resp = httpx.post(
+                    f"{base_url}/chat/completions",
+                    headers={
+                        "Authorization": f"Bearer {api_key}",
+                        "Content-Type": "application/json",
+                    },
+                    json={
+                        "model": model,
+                        "stream": False,
+                        "max_tokens": 1,
+                        "messages": [{"role": "user", "content": "ping"}],
+                    },
+                    timeout=timeout,
+                )
+                if resp.status_code == 200:
+                    logger.debug("Z.AI endpoint probe: %s (%s) model=%s OK", ep_id, base_url, model)
+                    return {
+                        "id": ep_id,
+                        "base_url": base_url,
+                        "model": model,
+                        "label": label,
+                    }
+                logger.debug("Z.AI endpoint probe: %s model=%s returned %s", ep_id, model, resp.status_code)
+            except Exception as exc:
+                logger.debug("Z.AI endpoint probe: %s model=%s failed: %s", ep_id, model, exc)
     return None

hermes_cli/models.py

@@ -44,6 +44,7 @@
     ("minimax/minimax-m2.7",            ""),
     ("minimax/minimax-m2.5",            ""),
     ("z-ai/glm-5.1",                    ""),
+    ("z-ai/glm-5v-turbo",               ""),
     ("z-ai/glm-5-turbo",                ""),
     ("moonshotai/kimi-k2.5",            ""),
     ("x-ai/grok-4.20",                  ""),
@@ -89,6 +90,7 @@ def _codex_curated_models() -> list[str]:
         "minimax/minimax-m2.7",
         "minimax/minimax-m2.5",
         "z-ai/glm-5.1",
+        "z-ai/glm-5v-turbo",
         "z-ai/glm-5-turbo",
         "moonshotai/kimi-k2.5",
         "x-ai/grok-4.20-beta",
@@ -134,6 +136,7 @@ def _codex_curated_models() -> list[str]:
     "zai": [
         "glm-5.1",
         "glm-5",
+        "glm-5v-turbo",
         "glm-5-turbo",
         "glm-4.7",
         "glm-4.5",

10 PRs closed without merging

From the close comments: superseded by another PR 4 · done elsewhere 2 · merged via another change 1 · other 3

#967 feat(providers): add Kimi Coding endpoint with auto-detection · merged via another change
#5566 fix(credential_pool): use _resolve_kimi_base_url when seeding kimi-coding pool · other → #5669
#5669 fix(auth): kimi-coding pool base_url seeding + PKCE endpoint fallback · other
#13954 fix(setup): clarify Moonshot/Kimi provider labels and accept MOONSHOT_API_KEY (#13935) · other
#42678 fix(auth): reorder Z.AI endpoint probe to try coding plan first (#42536) · superseded by another PR → #42547, #42536

Show 5 more

#43629 fix(provider): use api.kimi.com/coding endpoint for kimi-coding provider · done elsewhere → #43617
#43813 fix(provider): update kimi-coding endpoint and User-Agent (#43617) · superseded by another PR → #43629, #43617
#48509 fix(auxiliary_client): resolve API keys for inferred custom endpoints · done elsewhere → a8841e2
#58088 fix(zai): honor configured base url · superseded by another PR → 23d3eda
#61575 fix(agent): route Z.AI manual-pool entries through _resolve_zai_base_url · superseded by another PR

Closed issues · 8

Closed#966[Feature]: Add Kimi Coding as a direct inference provider
Closed#5561kimi-coding credential pool seeds wrong base_url for sk-kimi- keys, causing 401 on first request
Closed#9888[Feature]: Glm-5v-turbo for GLM Coding Plan
Closed#13935[Bug/UX]: kimi-coding provider misleading — points to kimi.com/coding not platform.moonshot.ai
Closed#43617[Bug]: kimi-coding provider uses wrong endpoint + User-Agent for sk-kimi-* keys, breaking all API calls including vision
Closed#58071[Bug] zai provider ignores config.yaml base_url and always probes endpoints in hardcoded order — standard endpoint always wins over coding endpoint
Closed#61563zai manual-pool credentials always routed to metered /api/paas/v4 instead of /api/coding/paas/v4 (hermes auth add cannot override base_url)
Closed#68239hermes auth add stamps registry-default base_url for kimi-coding — sk-kimi- keys paired with api.moonshot.ai/v1, instant 401

How this cluster was grouped

Link evidence (partly structural links): some members are connected by structural evidence (shared changed files, or a closing/superseded reference); the rest are joined by an automated relevance judgement over their text.
Members outside that structural core may belong to a different underlying problem.
Pairwise link density 0.05: the share of item pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36. 1.00 means every pair is linked.

Pick up a Lab task

Check whether a fix covers the open issues here, with your own agent. Bring back evidence for maintainers.

Open the Lab

At a glance

Subsystem: agent
Merged PRs: #9907
Closed issues: 8
Attempts: 10 PRs closed without merging
Fix link checked: 14 Apr 2026. This confirmation is old and may be out of date (a revert or reopen would not have invalidated it here); re-verify.

For your agent

The same facts, typed.

cluster/966.json
cluster/966.md