smashimo/oh-my-pi
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(catalog): strip stale xAI Responses effort dials from generated rows

Browse Source 
Paid xAI models.dev regeneration still emitted Completions-era thinking
dials for off-allowlist reasoners. Bake the no-dial policy into the
resolver/generator and refresh the exported catalog snapshot.
This commit is contained in:
Yang Yang
2026-08-02 22:47:15 -07:00
parent 3fb57803f3
commit b49b5b88d2
8 changed files with 479 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/ai/CHANGELOG.md
+4 -3
View File
@@ -2,6 +2,10 @@
## [Unreleased]
### Fixed
- Stopped treating `XAI_API_KEY` as SuperGrok (`xai-oauth`) sign-in for availability, so paid-key-only setups default to `xai/grok-4.5` instead of the zero-cost SuperGrok catalog path.
## [17.3.4] - 2026-08-14
### Fixed
@@ -148,9 +152,6 @@
### Fixed
- Fixed an issue where Ollama requests without a user-role message would fail to generate output or silently fail with a misleading error.
### Fixed
- Stopped treating `XAI_API_KEY` as SuperGrok (`xai-oauth`) sign-in for availability, so paid-key-only setups default to `xai/grok-4.5` instead of the zero-cost SuperGrok catalog path.
## [17.2.5] - 2026-08-03
packages/catalog/CHANGELOG.md
+8 -15
View File
@@ -8,6 +8,14 @@
- Changed the paid xAI (`XAI_API_KEY`) default model from `grok-4-fast-non-reasoning` to `grok-4.5`.
- Changed the SuperGrok (`xai-oauth`) default model from `grok-4.3` to `grok-4.5`.
- Requested `reasoning.encrypted_content` on first-party xAI Responses calls (`xai` and `xai-oauth`) via the `include` parameter.
- Replayed xAI encrypted reasoning items on later Responses turns instead of stripping `type: "reasoning"` history.
### Fixed
- Invalidated stale paid-xAI model-cache rows written under Chat Completions so the Responses migration takes effect immediately instead of waiting for TTL expiry.
- Clamped paid xAI Responses `minimal` reasoning effort to `low` (same wire map as SuperGrok) so `xai/grok-4.5` does not 400.
- Suppressed presence/frequency penalties and stop sequences on xAI reasoning models so a configured `presencePenalty` does not 400 after the `grok-4.5` default change.
- Stopped emitting stale `thinking.efforts` dials on paid xAI Responses catalog rows that reject `reasoning.effort` (`grok-code-fast-1`, `grok-build-0.1`, `grok-4.20-0309-reasoning`, and other off-allowlist reasoners).
## [17.3.4] - 2026-08-14
@@ -134,21 +142,6 @@
- Fixed dynamic discovery for the `deepseek-v4` model family (such as `deepseek-v4-flash-0731`) under `alibaba-token-plan` missing reasoning configuration and maximum thinking effort.
- Fixed GitHub Copilot dynamic discovery retaining stale bundled prices for default-context models instead of using the provider's reported default-tier prices.
## [17.2.5] - 2026-08-03
### Changed
- Switched the paid xAI provider (`xai` / `XAI_API_KEY`) from Chat Completions to the OpenAI Responses API (`POST https://api.x.ai/v1/responses`), matching SuperGrok `xai-oauth`. Prompt-cache affinity (`x-grok-conv-id`), reasoning-effort allowlisting, and encrypted-reasoning replay rules are now shared across both first-party xAI hosts.
- Changed the paid xAI (`XAI_API_KEY`) default model from `grok-4-fast-non-reasoning` to `grok-4.5`.
- Changed the SuperGrok (`xai-oauth`) default model from `grok-4.3` to `grok-4.5`.
- Requested `reasoning.encrypted_content` on first-party xAI Responses calls (`xai` and `xai-oauth`) via the `include` parameter.
- Replayed xAI encrypted reasoning items on later Responses turns instead of stripping `type: "reasoning"` history.
### Fixed
- Invalidated stale paid-xAI model-cache rows written under Chat Completions so the Responses migration takes effect immediately instead of waiting for TTL expiry.
- Clamped paid xAI Responses `minimal` reasoning effort to `low` (same wire map as SuperGrok) so `xai/grok-4.5` does not 400.
- Suppressed presence/frequency penalties and stop sequences on xAI reasoning models so a configured `presencePenalty` does not 400 after the `grok-4.5` default change.
## [17.2.5] - 2026-08-03
### Fixed
packages/catalog/scripts/generated-policies.ts
+5
View File
@@ -22,6 +22,7 @@ import { isOllamaCloudOutputCapped, OLLAMA_CLOUD_MAX_OUTPUT_TOKENS } from "../sr
import {
ALIBABA_TOKEN_PLAN_STATIC_MODELS,
OPENAI_GPT_56_LONG_CONTEXT_COSTS,
applyXaiResponsesThinkingPolicy,
resolveWaferServerlessThinkingFormat,
} from "../src/provider-models/openai-compat";
import type { Api, LongContextTokenCost, Model, ModelSpec } from "../src/types";
@@ -353,6 +354,10 @@ export function applyOllamaCloudOutputCap(models: ModelSpec<Api>[]): void {
}
function applyGeneratedModelPolicy(model: ModelSpec<Api>): void {
if (model.provider === "xai" && model.api === "openai-responses") {
const updated = applyXaiResponsesThinkingPolicy(model as ModelSpec<"openai-responses">);
model.compat = updated.compat;
}
const copilotLimits = model.provider === "github-copilot" ? COPILOT_GENERATED_LIMITS[model.id] : undefined;
if (copilotLimits) {
model.contextWindow = copilotLimits.contextWindow;
packages/catalog/src/models.json
+261 -88
View File
@@ -103730,7 +103730,14 @@
"cacheWrite": 0
},
"contextWindow": 131072,
"maxTokens": 8192
"maxTokens": 8192,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-2-1212": {
"id": "grok-2-1212",
@@ -103749,7 +103756,14 @@
"cacheWrite": 0
},
"contextWindow": 131072,
"maxTokens": 8192
"maxTokens": 8192,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-2-latest": {
"id": "grok-2-latest",
@@ -103768,7 +103782,14 @@
"cacheWrite": 0
},
"contextWindow": 131072,
"maxTokens": 8192
"maxTokens": 8192,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-2-vision": {
"id": "grok-2-vision",
@@ -103788,7 +103809,14 @@
"cacheWrite": 0
},
"contextWindow": 8192,
"maxTokens": 4096
"maxTokens": 4096,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-2-vision-1212": {
"id": "grok-2-vision-1212",
@@ -103808,7 +103836,14 @@
"cacheWrite": 0
},
"contextWindow": 8192,
"maxTokens": 4096
"maxTokens": 4096,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-2-vision-latest": {
"id": "grok-2-vision-latest",
@@ -103828,7 +103863,14 @@
"cacheWrite": 0
},
"contextWindow": 8192,
"maxTokens": 4096
"maxTokens": 4096,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-3": {
"id": "grok-3",
@@ -103847,7 +103889,14 @@
"cacheWrite": 0
},
"contextWindow": 131072,
"maxTokens": 8192
"maxTokens": 8192,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-3-fast": {
"id": "grok-3-fast",
@@ -103866,7 +103915,14 @@
"cacheWrite": 0
},
"contextWindow": 131072,
"maxTokens": 8192
"maxTokens": 8192,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-3-fast-latest": {
"id": "grok-3-fast-latest",
@@ -103885,7 +103941,14 @@
"cacheWrite": 0
},
"contextWindow": 131072,
"maxTokens": 8192
"maxTokens": 8192,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-3-latest": {
"id": "grok-3-latest",
@@ -103904,7 +103967,14 @@
"cacheWrite": 0
},
"contextWindow": 131072,
"maxTokens": 8192
"maxTokens": 8192,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-3-mini": {
"id": "grok-3-mini",
@@ -103930,8 +104000,19 @@
"minimal",
"low",
"medium",
"high"
]
"high",
"xhigh"
],
"effortMap": {
"minimal": "low"
}
},
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": true,
"omitReasoningEffort": false
}
},
"grok-3-mini-fast": {
@@ -103958,8 +104039,19 @@
"minimal",
"low",
"medium",
"high"
]
"high",
"xhigh"
],
"effortMap": {
"minimal": "low"
}
},
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": true,
"omitReasoningEffort": false
}
},
"grok-3-mini-fast-latest": {
@@ -103986,8 +104078,19 @@
"minimal",
"low",
"medium",
"high"
]
"high",
"xhigh"
],
"effortMap": {
"minimal": "low"
}
},
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": true,
"omitReasoningEffort": false
}
},
"grok-3-mini-latest": {
@@ -104014,8 +104117,19 @@
"minimal",
"low",
"medium",
"high"
]
"high",
"xhigh"
],
"effortMap": {
"minimal": "low"
}
},
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": true,
"omitReasoningEffort": false
}
},
"grok-4": {
@@ -104036,14 +104150,12 @@
},
"contextWindow": 256000,
"maxTokens": 64000,
"thinking": {
"mode": "effort",
"efforts": [
"minimal",
"low",
"medium",
"high"
]
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-4-1-fast": {
@@ -104065,14 +104177,12 @@
},
"contextWindow": 2000000,
"maxTokens": 30000,
"thinking": {
"mode": "effort",
"efforts": [
"minimal",
"low",
"medium",
"high"
]
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-4-1-fast-non-reasoning": {
@@ -104093,7 +104203,14 @@
"cacheWrite": 0
},
"contextWindow": 2000000,
"maxTokens": 30000
"maxTokens": 30000,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-4-fast": {
"id": "grok-4-fast",
@@ -104114,14 +104231,12 @@
},
"contextWindow": 2000000,
"maxTokens": 30000,
"thinking": {
"mode": "effort",
"efforts": [
"minimal",
"low",
"medium",
"high"
]
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-4-fast-non-reasoning": {
@@ -104142,7 +104257,14 @@
"cacheWrite": 0
},
"contextWindow": 2000000,
"maxTokens": 30000
"maxTokens": 30000,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-4.20-0309-non-reasoning": {
"id": "grok-4.20-0309-non-reasoning",
@@ -104162,7 +104284,14 @@
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 30000
"maxTokens": 30000,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-4.20-0309-reasoning": {
"id": "grok-4.20-0309-reasoning",
@@ -104183,15 +104312,12 @@
},
"contextWindow": 1000000,
"maxTokens": 30000,
"thinking": {
"mode": "effort",
"efforts": [
"minimal",
"low",
"medium",
"high"
],
"requiresEffort": true
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-4.20-beta-latest-non-reasoning": {
@@ -104212,7 +104338,14 @@
"cacheWrite": 0
},
"contextWindow": 2000000,
"maxTokens": 30000
"maxTokens": 30000,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-4.20-beta-latest-reasoning": {
"id": "grok-4.20-beta-latest-reasoning",
@@ -104233,15 +104366,12 @@
},
"contextWindow": 2000000,
"maxTokens": 30000,
"thinking": {
"mode": "effort",
"efforts": [
"minimal",
"low",
"medium",
"high"
],
"requiresEffort": true
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-4.20-multi-agent-beta-latest": {
@@ -104269,8 +104399,19 @@
"minimal",
"low",
"medium",
"high"
]
"high",
"xhigh"
],
"effortMap": {
"minimal": "low"
}
},
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": true,
"omitReasoningEffort": false
}
},
"grok-4.3": {
@@ -104298,8 +104439,19 @@
"minimal",
"low",
"medium",
"high"
]
"high",
"xhigh"
],
"effortMap": {
"minimal": "low"
}
},
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": true,
"omitReasoningEffort": false
}
},
"grok-4.5": {
@@ -104327,8 +104479,19 @@
"minimal",
"low",
"medium",
"high"
]
"high",
"xhigh"
],
"effortMap": {
"minimal": "low"
}
},
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": true,
"omitReasoningEffort": false
}
},
"grok-4.6": {
@@ -104377,7 +104540,14 @@
"cacheWrite": 0
},
"contextWindow": 131072,
"maxTokens": 4096
"maxTokens": 4096,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-build-0.1": {
"id": "grok-build-0.1",
@@ -104398,14 +104568,12 @@
},
"contextWindow": 256000,
"maxTokens": 256000,
"thinking": {
"mode": "effort",
"efforts": [
"minimal",
"low",
"medium",
"high"
]
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-code-fast-1": {
@@ -104426,14 +104594,12 @@
},
"contextWindow": 256000,
"maxTokens": 10000,
"thinking": {
"mode": "effort",
"efforts": [
"minimal",
"low",
"medium",
"high"
]
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
},
"grok-vision-beta": {
@@ -104454,7 +104620,14 @@
"cacheWrite": 0
},
"contextWindow": 8192,
"maxTokens": 4096
"maxTokens": 4096,
"compat": {
"reasoningEffortMap": {
"minimal": "low"
},
"supportsReasoningEffort": false,
"omitReasoningEffort": true
}
}
},
"xai-oauth": {
packages/catalog/src/provider-models/openai-compat.ts
+33 -1
View File
@@ -1376,6 +1376,36 @@ function withXaiOAuthCompatDefaults(model: ModelSpec<"openai-responses">): Model
// of the omitReasoningEffort gate in pi-ai's stream.ts.
const XAI_REASONING_EFFORT_MAP = { minimal: "low" } as const;
/**
* Bake first-party xAI Responses effort-dial metadata onto a catalog spec.
*
* models.dev marks many Grok SKUs as reasoners and the thinking rebake would
* otherwise emit a default `minimal/low/medium/high` dial. api.x.ai only
* accepts `reasoning.effort` for {@link isGrokReasoningEffortCapable} ids —
* off-allowlist reasoners (`grok-code-fast-1`, `grok-build-0.1`,
* `grok-4.20-0309-reasoning`, …) 400 if the param is sent. SuperGrok
* (`xai-oauth`) already curates this via {@link mergeCuratedIntoModel}; paid
* `xai` rows come from stencil.so and need the same wire facts in the exported
* `models.json` so direct catalog readers do not present an unsupported dial.
*
* Explicit `compat.supportsReasoningEffort` / `omitReasoningEffort` win.
*/
export function applyXaiResponsesThinkingPolicy(model: ModelSpec<"openai-responses">): ModelSpec<"openai-responses"> {
const effortCapable = model.compat?.supportsReasoningEffort ?? isGrokReasoningEffortCapable(model.id);
return {
...model,
compat: {
...(model.compat ?? {}),
reasoningEffortMap: {
...XAI_REASONING_EFFORT_MAP,
...(model.compat?.reasoningEffortMap ?? {}),
},
supportsReasoningEffort: effortCapable,
omitReasoningEffort: model.compat?.omitReasoningEffort ?? !effortCapable,
},
};
}
// xai-oauth's /v1/models exposes no per-request output limit on the OAuth
// (Grok Build / SuperGrok) surface, so the curated catalog owns `maxTokens`
// like it owns `contextWindow`: each entry mirrors its context window. The
@@ -5854,7 +5884,9 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_CORE: readonly ModelsDevProviderDescriptor
defaultContextWindow: 131072,
}),
// --- xAI ---
openAiResponsesDescriptor("xai", "xai", "https://api.x.ai/v1"),
openAiResponsesDescriptor("xai", "xai", "https://api.x.ai/v1", {
transformModel: model => applyXaiResponsesThinkingPolicy(model as ModelSpec<"openai-responses">),
}),
// --- DeepSeek ---
openAiCompletionsDescriptor("deepseek", "deepseek", "https://api.deepseek.com", {
// Only ship the v4 family as built-ins; older deepseek-chat / deepseek-reasoner
packages/catalog/test/generated-policies.test.ts
+43
View File
@@ -467,6 +467,49 @@ describe("generated model policies", () => {
expect(models[2]?.applyPatchToolType).toBeUndefined();
expect(models[3]?.applyPatchToolType).toBeUndefined();
});
it("strips paid xAI Responses effort dials for off-allowlist reasoners", () => {
const models: ModelSpec<"openai-responses">[] = [
createSpec({
id: "grok-code-fast-1",
api: "openai-responses",
provider: "xai",
thinking: { mode: "effort", efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High] },
}),
createSpec({
id: "grok-4.5",
api: "openai-responses",
provider: "xai",
thinking: { mode: "effort", efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High] },
}),
createSpec({
id: "grok-code-fast-1",
api: "openai-responses",
provider: "openrouter",
thinking: { mode: "effort", efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High] },
}),
];
applyGeneratedModelPolicies(models);
expect(models[0]?.thinking).toBeUndefined();
expect(models[0]?.compat).toMatchObject({
supportsReasoningEffort: false,
omitReasoningEffort: true,
reasoningEffortMap: { minimal: "low" },
});
expect(models[1]?.thinking?.efforts).toEqual([
Effort.Minimal,
Effort.Low,
Effort.Medium,
Effort.High,
Effort.XHigh,
]);
expect(models[1]?.compat?.supportsReasoningEffort).toBe(true);
// Non-xAI hosts are outside this policy — no baked no-dial compat.
expect(models[2]?.thinking).toBeDefined();
expect(models[2]?.compat?.supportsReasoningEffort).toBeUndefined();
});
});
describe("applyOllamaCloudOutputCap", () => {
packages/catalog/test/xai-responses-thinking-policy.test.ts
+122
View File
@@ -0,0 +1,122 @@
import { describe, expect, it } from "bun:test";
import { Effort } from "@oh-my-pi/pi-catalog/effort";
import MODELS_JSON from "@oh-my-pi/pi-catalog/models.json" with { type: "json" };
import {
MODELS_DEV_PROVIDER_DESCRIPTORS,
mapModelsDevToModels,
} from "@oh-my-pi/pi-catalog/provider-models/openai-compat";
import type { ModelSpec } from "@oh-my-pi/pi-catalog/types";
import { applyGeneratedModelPolicies } from "../scripts/generated-policies";
const XAI_MODELS_DEV_FIXTURE = {
xai: {
models: {
"grok-4.5": {
name: "Grok 4.5",
tool_call: true,
reasoning: true,
modalities: { input: ["text", "image"] },
limit: { context: 500_000, output: 500_000 },
cost: { input: 2, output: 6, cache_read: 0.3 },
},
"grok-code-fast-1": {
name: "Grok Code Fast 1",
tool_call: true,
reasoning: true,
modalities: { input: ["text"] },
limit: { context: 256_000, output: 10_000 },
cost: { input: 0.2, output: 1.5 },
},
"grok-build-0.1": {
name: "Grok Build 0.1",
tool_call: true,
reasoning: true,
modalities: { input: ["text", "image"] },
limit: { context: 256_000, output: 256_000 },
cost: { input: 0, output: 0 },
},
"grok-4.20-0309-reasoning": {
name: "Grok 4.20 (Reasoning)",
tool_call: true,
reasoning: true,
modalities: { input: ["text", "image"] },
limit: { context: 2_000_000, output: 64_000 },
cost: { input: 2, output: 6 },
},
"grok-2": {
name: "Grok 2",
tool_call: true,
reasoning: false,
modalities: { input: ["text"] },
limit: { context: 131_072, output: 8192 },
cost: { input: 2, output: 10 },
},
},
},
};
describe("paid xAI Responses thinking policy", () => {
it("bakes the effort-dial allowlist on stencil.so → openai-responses mapping", () => {
const mapped = mapModelsDevToModels(XAI_MODELS_DEV_FIXTURE, MODELS_DEV_PROVIDER_DESCRIPTORS).filter(
model => model.provider === "xai",
);
const byId = Object.fromEntries(mapped.map(model => [model.id, model]));
expect(byId["grok-4.5"]?.api).toBe("openai-responses");
expect(byId["grok-4.5"]?.compat).toMatchObject({
supportsReasoningEffort: true,
omitReasoningEffort: false,
reasoningEffortMap: { minimal: "low" },
});
for (const id of ["grok-code-fast-1", "grok-build-0.1", "grok-4.20-0309-reasoning"] as const) {
expect(byId[id]?.reasoning, id).toBe(true);
expect(byId[id]?.compat, id).toMatchObject({
supportsReasoningEffort: false,
omitReasoningEffort: true,
reasoningEffortMap: { minimal: "low" },
});
}
expect(byId["grok-2"]?.compat).toMatchObject({
supportsReasoningEffort: false,
omitReasoningEffort: true,
});
});
it("strips stale thinking dials from off-allowlist paid xAI reasoners during generation", () => {
const mapped = mapModelsDevToModels(XAI_MODELS_DEV_FIXTURE, MODELS_DEV_PROVIDER_DESCRIPTORS).filter(
model => model.provider === "xai",
);
// Snapshot-era Completions rows still carry a default effort ladder after the
// api flip; the generator must not re-emit that dial for Responses.
const snapshotStale = mapped.find(model => model.id === "grok-code-fast-1");
expect(snapshotStale).toBeDefined();
snapshotStale!.thinking = { mode: "effort", efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High] };
applyGeneratedModelPolicies(mapped);
const byId = Object.fromEntries(mapped.map(model => [model.id, model]));
expect(byId["grok-4.5"]?.thinking).toEqual({
mode: "effort",
efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh],
effortMap: { minimal: "low" },
});
for (const id of ["grok-code-fast-1", "grok-build-0.1", "grok-4.20-0309-reasoning"] as const) {
expect(byId[id]?.reasoning, id).toBe(true);
expect(byId[id]?.thinking, id).toBeUndefined();
expect(byId[id]?.compat, id).toMatchObject({ supportsReasoningEffort: false });
}
});
it("exports no-dial rows in the bundled models.json snapshot", () => {
const bundled =
(MODELS_JSON as unknown as Record<string, Record<string, ModelSpec<"openai-responses">>>).xai ?? {};
for (const id of ["grok-code-fast-1", "grok-build-0.1", "grok-4.20-0309-reasoning"] as const) {
expect(bundled[id], `xai/${id} missing from models.json`).toBeDefined();
expect(bundled[id]?.reasoning, id).toBe(true);
expect(bundled[id]?.thinking, id).toBeUndefined();
expect(bundled[id]?.compat?.supportsReasoningEffort, id).toBe(false);
}
expect(bundled["grok-4.5"]?.thinking?.efforts).toContain(Effort.XHigh);
expect(bundled["grok-4.5"]?.compat?.supportsReasoningEffort).toBe(true);
});
});
packages/coding-agent/CHANGELOG.md
+3 -9
View File
@@ -8,6 +8,9 @@
- Changed the default model for `XAI_API_KEY` (`xai`) from `grok-4-fast-non-reasoning` to `grok-4.5`.
- Changed the default model for SuperGrok OAuth (`xai-oauth`) from `grok-4.3` to `grok-4.5`.
- Included `reasoning.encrypted_content` in Responses `include` for paid xAI and SuperGrok OAuth models.
- Replayed encrypted xAI reasoning on follow-up Responses turns for `xai` and `xai-oauth`.
- Kept automatic model selection on paid `xai/grok-4.5` when only `XAI_API_KEY` is set, instead of preferring SuperGrok `xai-oauth/grok-4.5`.
- Stopped sending presence/frequency penalties and stop sequences to xAI reasoning models such as `grok-4.5`, which reject them.
## [17.3.4] - 2026-08-14
@@ -364,15 +367,6 @@
- Fixed issues with `/btw` branch promotion where branches could park behind active turns, cut from outdated session leaves, or leave rejected branch keys indistinguishable from composer input.
- Fixed database bloat by ensuring archived main and nested session rows are properly cleaned up from `stats.db` during garbage collection.
- Fixed startup hanging during local model discovery when a timed-out transport left its request pending, which blocked the CLI before OAuth login could finish ([#7482](https://github.com/can1357/oh-my-pi/issues/7482)).
### Changed
- Routed paid xAI models (`XAI_API_KEY` / `xai/…`) through the Responses API used by SuperGrok OAuth instead of Chat Completions.
- Changed the default model for `XAI_API_KEY` (`xai`) from `grok-4-fast-non-reasoning` to `grok-4.5`.
- Changed the default model for SuperGrok OAuth (`xai-oauth`) from `grok-4.3` to `grok-4.5`.
- Included `reasoning.encrypted_content` in Responses `include` for paid xAI and SuperGrok OAuth models.
- Replayed encrypted xAI reasoning on follow-up Responses turns for `xai` and `xai-oauth`.
- Kept automatic model selection on paid `xai/grok-4.5` when only `XAI_API_KEY` is set, instead of preferring SuperGrok `xai-oauth/grok-4.5`.
- Stopped sending presence/frequency penalties and stop sequences to xAI reasoning models such as `grok-4.5`, which reject them.
## [17.2.5] - 2026-08-03