Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 8 additions & 6 deletions docs/config/models.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -18,9 +18,9 @@ Xum ships with curated models kept up to date with the frontier. Use any custom
| Opus 5.5 | anthropic:claude-opus-5-5 | `opus` | ✓ |
| Sonnet 5.5 | anthropic:claude-sonnet-5-5 | `sonnet` | |
| Haiku 5.5 | anthropic:claude-haiku-5-5 | `haiku` | |
| GPT-6.1 Sol | openai:gpt-6.1-sol | `gpt`, `sol` | |
| GPT-6.2 Sol | openai:gpt-6.2-sol | `gpt`, `sol` | |
| GPT-6 Luna | openai:gpt-6-luna | `luna` | |
| GPT-6 Astra | openai:gpt-6-astra | `astra`, `gpt-6-astra` | |
| GPT-6.2 Astra | openai:gpt-6.2-astra | `astra`, `gpt-6.2-astra` | |
| Gemini 3.1 Pro Preview | google:gemini-3.1-pro-preview | `gemini`, `gemini-pro` | |
| Gemini 3.8 Flash | google:gemini-3.8-flash | `gemini-flash` | |
| Gemini 4 Argon | google:gemini-4-argon | `argon`, `gemini-argon`, `gemini-4` | |
Expand All @@ -34,21 +34,23 @@ Xum ships with curated models kept up to date with the frontier. Use any custom

{/* END KNOWN_MODELS_TABLE */}

Xum supports the GPT-6 tiers (GPT-6.1 Sol, GPT-6 Astra, and GPT-6 Luna); older OpenAI GPT models are no longer supported.
Xum supports the GPT-6 tiers (GPT-6.2 Sol, GPT-6.2 Astra, and GPT-6 Luna); older OpenAI GPT models are no longer supported. GPT-6.1 Sol and GPT-6 Astra stay usable as the custom model strings `openai:gpt-6.1-sol` and `openai:gpt-6-astra`.

GPT-5.6 models added from **Settings → Models** (`gpt-5.6`, `gpt-5.6-sol`, `gpt-5.6-terra`, and `gpt-5.6-luna`) support thinking from off through max and Pro reasoning mode. Other older GPT models stay unsupported.

GPT-6 Luna supports thinking from off through max on the default Responses API route. On the opt-in Chat Completions route, Xum disables its reasoning because OpenAI only supports function calling there with reasoning off.

GPT-6.1 Sol (the `gpt` and `sol` aliases) and GPT-6 Astra support thinking from low through max; OpenAI does not allow turning their reasoning off. Use the default Responses API route with them, because OpenAI does not support tool calling for these models on Chat Completions.
GPT-6.2 Sol (the `gpt` and `sol` aliases), GPT-6.2 Astra (the `astra` alias), GPT-6.1 Sol, and GPT-6 Astra support thinking from low through max; OpenAI does not allow turning their reasoning off. Use the default Responses API route with them, because OpenAI does not support tool calling for these models on Chat Completions.

### OpenAI Ultrafast

To use OpenAI's fastest processing tier, turn on the thinking selector's **Ultrafast mode** row (it sits next to **Fast mode**), run **Toggle Ultrafast Mode** from the command palette, or use its keyboard shortcut. You can also choose **ultrafast** as the service tier in **Settings → Providers → OpenAI**. Fast and Ultrafast are mutually exclusive, and turning either off switches the chat to Standard. Like the thinking level, the choice is saved per chat and stays with the chat when you switch models; chats without a choice use the provider's service tier. Xum sends it only to models that OpenAI serves Ultrafast for: GPT-6.1 Sol and GPT-6 Astra. It also needs a route that sends OpenAI's own Responses API: direct OpenAI on the default wire format, Coder instances of type OpenAI, or custom OpenAI Responses providers. Direct OpenAI set to Chat Completions, Coder OpenAI-compatible instances, OpenRouter, and GitHub Copilot do not accept the tier. Elsewhere the request runs without a service tier instead of falling back to Fast. Ultrafast costs 6× the Standard price, and Xum's cost tracking uses that rate.
To use OpenAI's fastest processing tier, turn on the thinking selector's **Ultrafast mode** row (it sits next to **Fast mode**), run **Toggle Ultrafast Mode** from the command palette, or use its keyboard shortcut. You can also choose **ultrafast** as the service tier in **Settings → Providers → OpenAI**. Fast and Ultrafast are mutually exclusive, and turning either off switches the chat to Standard. Like the thinking level, the choice is saved per chat and stays with the chat when you switch models; chats without a choice use the provider's service tier. Xum sends it only to models that OpenAI serves Ultrafast for: GPT-6.2 Sol, GPT-6.2 Astra, GPT-6.1 Sol, and GPT-6 Astra. It also needs a route that sends OpenAI's own Responses API: direct OpenAI on the default wire format, Coder instances of type OpenAI, or custom OpenAI Responses providers. Direct OpenAI set to Chat Completions, Coder OpenAI-compatible instances, OpenRouter, and GitHub Copilot do not accept the tier. Elsewhere the request runs without a service tier instead of falling back to Fast. Ultrafast costs 6× the Standard price, and Xum's cost tracking uses that rate. OpenAI charges 3× for GPT-6.2 Sol and Astra through February 28, 2027, so until then Xum shows twice their real Ultrafast cost.

OpenAI does not offer the Flex tier for GPT-6.2 Sol and Astra. If **Settings → Providers → OpenAI** sets the service tier to **flex**, their requests run at Standard instead.

### Pro reasoning mode

GPT-6.1 Sol, GPT-6 Astra, GPT-6 Luna, and the GPT-5.6 models added from **Settings → Models** support OpenAI's pro reasoning mode. Open the reasoning selector next to the model picker and enable **Pro mode** (or run "Toggle Pro Reasoning Mode" from the Command Palette) to send `reasoning.mode: "pro"` with each request. The setting is saved per workspace. Pro mode can consume more tokens and responses can take noticeably longer. Pro is available on supported direct OpenAI Responses API routes. Gateway routes, Chat Completions, and Codex OAuth use standard mode.
The GPT-6 tiers (GPT-6.2 Sol, GPT-6.2 Astra, GPT-6.1 Sol, GPT-6 Astra, and GPT-6 Luna) and the GPT-5.6 models added from **Settings → Models** support OpenAI's pro reasoning mode. Open the reasoning selector next to the model picker and enable **Pro mode** (or run "Toggle Pro Reasoning Mode" from the Command Palette) to send `reasoning.mode: "pro"` with each request. The setting is saved per workspace. Pro mode can consume more tokens and responses can take noticeably longer. Pro is available on supported direct OpenAI Responses API routes. Gateway routes, Chat Completions, and Codex OAuth use standard mode.

## Model Selection

Expand Down
10 changes: 6 additions & 4 deletions docs/config/providers.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -167,8 +167,8 @@ The built-in `openai` provider supports two wire formats:
| `responses` | Uses the OpenAI Responses API. This is the default and supports current OpenAI and Codex features. |
| `chatCompletions` | Uses `/v1/chat/completions`. Use this for OpenAI-API-shaped gateways that do not expose `/v1/responses`, such as Azure Government endpoints. |

OpenAI documents that GPT-6 Astra only supports tool calling through the Responses API, so keep
`responses` when using that model.
OpenAI documents that GPT-6 Astra and the GPT-6.2 models only support tool calling through the
Responses API, so keep `responses` when using them.
Comment thread
ThomasK33 marked this conversation as resolved.
Resolved

Set the format in **Settings → Providers → OpenAI → Wire format**, or in
`~/.xum/providers.jsonc`:
Expand Down Expand Up @@ -200,8 +200,10 @@ Palette. Cyber and Pro are mutually exclusive.

Cyber appears only for models with a documented Daybreak value. GPT-6.1 Sol and GPT-6 Astra send
`daybreak_blue`. OpenAI requires Daybreak Red approval for reduced refusals on these models, but
they still take `daybreak_blue`. Cyber works only with the built-in OpenAI provider's API key on
the `responses` wire format. A custom base URL on the built-in provider also receives the field.
they still take `daybreak_blue`. OpenAI's guide lists no value for GPT-6.2 Sol or Astra, so Cyber
does not appear for them or for the `gpt`, `sol`, and `astra` aliases; to use Cyber, pick
`openai:gpt-6.1-sol` or `openai:gpt-6-astra`. Cyber works only with the built-in OpenAI
provider's API key on the `responses` wire format. A custom base URL on the built-in provider also receives the field.
Gateways, custom providers, Codex OAuth, and Chat Completions use standard safeguards.

If OpenAI rejects the program, Xum shows OpenAI's message and does not retry:
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -595,30 +595,30 @@
const saveBashAiProxyEnabled = (api: SettingsApi, enabled: boolean) =>
api.config.updateBashAiProxyEnabled({ enabled });

export function ProvidersSection() {
const {
providersExpandedProvider,
setProvidersExpandedProvider,
providersStartCoderLogin,
setProvidersStartCoderLogin,
} = useSettings();

// The live client only, for gates, disabled states and writes: offline (the last client kept
// while reconnecting) a write would fail, and the optimistic value would look saved until the
// reconnect puts the server value back. Loads go through useProvidersConfig.
const liveApi = useLiveAPI();
const api = liveApi;
const { config, loading: configLoading, refresh, updateOptimistically } = useProvidersConfig();
const { refreshWorkspaceMetadata } = useWorkspaceContext();
const visibleProviders = useMemo(() => getProviderIdsForUi(config), [config]);

const [openaiServiceTierSelectOverride, setOpenaiServiceTierSelectOverride] =
useState<OpenAIServiceTierSelectValue | null>(null);
const [xaiServiceTierSaving, setXAIServiceTierSaving] = useState(false);

Check warning on line 617 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: false -> true

No unit test failed when ProvidersSection had false -> true. Add a test that tells them apart, or say why they are equivalent.
// Persist OpenAI ZDR store toggles before publishing UI state so a failed write
// cannot leave the dropdown claiming disabled while requests still send store=true.
// xAI frontier Grok always uses store=false in the request path (no settings surface).
const [openaiStoreSaving, setOpenAIStoreSaving] = useState(false);

Check warning on line 621 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: false -> true

No unit test failed when ProvidersSection had false -> true. Add a test that tells them apart, or say why they are equivalent.
const [openaiCyberSaving, setOpenAICyberSaving] = useState(false);

const routing = useRouting();
Expand All @@ -627,7 +627,7 @@
// custom chat provider under the same id is listed with the custom providers instead.
const autoModelRoutingEnabled = useExperimentValue(EXPERIMENT_IDS.AUTO_MODEL_ROUTING);
const showTypeSafeProvider =
autoModelRoutingEnabled && !isCustomProviderInfo(config?.[TYPESAFE_PROVIDER_KEY]);

Check warning on line 630 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: delete ?.

No unit test failed when ProvidersSection had delete ?.. Add a test that tells them apart, or say why they are equivalent.

Check warning on line 630 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: && -> ||

No unit test failed when ProvidersSection had && -> ||. Add a test that tells them apart, or say why they are equivalent.

Check warning on line 630 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: delete !

No unit test failed when ProvidersSection had delete !. Add a test that tells them apart, or say why they are equivalent.

const providerGroups = useMemo(() => {
const groups: Record<"direct" | "gateway" | "local" | "custom", string[]> = {
Expand All @@ -637,14 +637,14 @@
custom: [],
};
for (const provider of visibleProviders) {
if (!isBuiltInProvider(provider) || isCustomProviderInfo(config?.[provider])) {

Check warning on line 640 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: || -> &&

No unit test failed when ProvidersSection had || -> &&. Add a test that tells them apart, or say why they are equivalent.

Check warning on line 640 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: delete ?.

No unit test failed when ProvidersSection had delete ?.. Add a test that tells them apart, or say why they are equivalent.
continue;
}

groups[PROVIDER_DEFINITIONS[provider].kind].push(provider);
}

for (const [provider, providerInfo] of Object.entries(config ?? {})) {

Check warning on line 647 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: ?? -> ||

No unit test failed when ProvidersSection had ?? -> ||. Add a test that tells them apart, or say why they are equivalent.
if (!isCustomProviderInfo(providerInfo)) {
continue;
}
Expand All @@ -657,7 +657,7 @@

const displayedRoutePriority = useMemo(() => {
const priority = routing.routePriority.filter(
(route) => route === "direct" || providerGroups.gateway.includes(route)

Check warning on line 660 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: || -> &&

No unit test failed when ProvidersSection had || -> &&. Add a test that tells them apart, or say why they are equivalent.

Check warning on line 660 in src/browser/features/Settings/Sections/ProvidersSection.tsx

View workflow job for this annotation

GitHub Actions / Quality / Mutation

Mutation survived: === -> !==

No unit test failed when ProvidersSection had === -> !==. Add a test that tells them apart, or say why they are equivalent.
);

if (!priority.includes("direct")) {
Expand Down Expand Up @@ -2544,7 +2544,7 @@
</label>
<p className="text-muted text-xs">
Applies to models that support both ChatGPT OAuth and API keys
(e.g. <code className="text-accent-text">gpt-6.1-sol</code>).
(e.g. <code className="text-accent-text">gpt-6.2-sol</code>).
</p>
</div>

Expand Down
3 changes: 2 additions & 1 deletion src/common/config/schemas/providersConfig.ts
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,8 @@ export const CacheTtlSchema = z.enum(["5m", "1h"]);
/** Anthropic request speed; "fast" opts into Fast mode (see anthropicFastMode.ts). */
export const AnthropicSpeedSchema = z.enum(["standard", "fast"]);
// "ultrafast" is OpenAI's model-gated premium tier (DevDay 2026-09-29); requests for
// models without it drop the tier (see openaiModelSupportsServiceTier).
// models without it drop the tier, as GPT-6.2 requests do for "flex" (see
// openaiModelSupportsServiceTier).
export const ServiceTierSchema = z.enum(["auto", "default", "flex", "priority", "ultrafast"]);
export type ServiceTier = z.infer<typeof ServiceTierSchema>;
// Fast mode remembers the tier it replaced, so toggling Fast off from Ultrafast restores it.
Expand Down
3 changes: 3 additions & 0 deletions src/common/constants/codexOAuth.test.ts
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,9 @@ describe("codexOAuth model gating", () => {
["gpt-6-astra", 272_000],
["gpt-6.1-sol", 272_000],
["gpt-6-luna", 272_000],
// No catalog row yet; the cap is assumed from the rows above.
["gpt-6.2-astra", 272_000],
["gpt-6.2-sol", 272_000],
])("allows %s through Codex OAuth with a %d context cap", (model, contextLimit) => {
for (const id of [model, `openai:${model}`]) {
expect(isCodexOauthAllowedModelId(id)).toBe(true);
Expand Down
16 changes: 14 additions & 2 deletions src/common/constants/codexOAuth.ts
Original file line number Diff line number Diff line change
Expand Up @@ -108,9 +108,15 @@ export const CODEX_OAUTH_ALLOWED_MODELS = new Set<string>([
// context-window overrides below), so the `luna` alias keeps working for
// OAuth-only users.
"gpt-6-luna",
// GPT-6.1 Sol (September 29, 2026): the Codex catalog's default model, so the
// promoted `gpt`/`sol` aliases keep working for OAuth-only users.
// GPT-6.1 Sol (September 29, 2026): the Codex catalog's default model when the
// `gpt`/`sol` aliases moved to it.
"gpt-6.1-sol",
// GPT-6.2 Astra and Sol (November 17, 2026): OpenAI announced both in Codex on
// release day, and the `astra` and `gpt`/`sol` aliases now point at them, so
// without these entries OAuth-only users of those aliases fail with
// api_key_not_found.
"gpt-6.2-astra",
"gpt-6.2-sol",
]);

/**
Expand All @@ -129,6 +135,12 @@ const CODEX_OAUTH_CONTEXT_WINDOW_OVERRIDES: Record<string, number> = {
"gpt-6-luna": 272_000,
// GPT-6.1 Sol: https://github.com/openai/codex/blob/b1e72963c3b71a9265a551e54beff078384efed9/codex-rs/models-manager/models.json
"gpt-6.1-sol": 272_000,
// GPT-6.2 Astra/Sol: ASSUMED, the one exception to the rule above. The catalog had
// no 6.2 row at c08843a5bd16de5524129b2f4270490447991984 (release day), so these
// copy the other GPT-6 rows' 272K. A cap below the real one only compacts
// earlier; replace them with the catalog's values once its rows land.
"gpt-6.2-astra": 272_000,
"gpt-6.2-sol": 272_000,
};

function normalizeCodexOauthModelId(modelId: string): string {
Expand Down
2 changes: 2 additions & 0 deletions src/common/constants/knownModels.test.ts
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,8 @@ describe("Known Models Integration", () => {
["anthropic:claude-haiku-4-5", "anthropic/claude-3.5-haiku"],
["anthropic:claude-fable-5-1", "anthropic/claude-opus-4.5"],
["anthropic:claude-mythos-5-1", "anthropic/claude-opus-4.5"],
["openai:gpt-6.1-sol", "openai/gpt-5"],
["openai:gpt-6-astra", "openai/gpt-5"],
])("retired id %s keeps its tokenizer override", (modelId, tokenizer) => {
expect(TOKENIZER_MODEL_OVERRIDES[modelId]).toBe(tokenizer);
});
Expand Down
46 changes: 19 additions & 27 deletions src/common/constants/knownModels.ts
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,6 @@
* Centralized model metadata. Update model versions here and everywhere else will follow.
*/

import type { OpenAICyberAccessProgram } from "../types/openaiAccessPrograms";
import { formatModelDisplayName } from "../utils/ai/modelDisplay";

type ModelProvider =
Expand Down Expand Up @@ -30,16 +29,6 @@ interface KnownModelDefinition {
warm?: boolean;
/** Optional tokenizer override for ai-tokenizer */
tokenizerOverride?: string;
/**
* OpenAI `access_programs.cyber` value sent in Cyber mode. Set only for the
* exact request model ids OpenAI's Daybreak guide maps to a program
* (https://developers.openai.com/api/docs/guides/daybreak, retrieved
* 2026-09-30). The value follows the model, not the organization's approval
* level. GPT-6 Luna is not listed, and GPT-6 Sol is listed but no longer a
* Xum model. Daybreak aliases and gpt-5.6-cyber accept only one program,
* which the API already selects when the field is omitted.
*/
cyberAccessProgram?: OpenAICyberAccessProgram;
}

interface KnownModel extends KnownModelDefinition {
Expand Down Expand Up @@ -128,22 +117,21 @@ const MODEL_DEFINITIONS = {
// counting (same situation as OPUS above).
tokenizerOverride: "anthropic/claude-opus-4.5",
},
// GPT-6.1 Sol - released September 29, 2026, successor to GPT-6 Sol at the same
// Standard pricing ($2/M input, $10/M output) with cheaper cache reads ($0.10/M).
// Like Astra, it cannot disable reasoning. Keep the durable gpt/sol aliases on the
// latest Sol tier without moving users to the more expensive Astra tier. OpenAI GPT
// models older than GPT-6 (and GPT-6 Sol) are no longer supported: Pro reasoning
// mode on the GPT-6 tiers supersedes the separate gpt-5.5-pro model id.
// GPT-6.2 Sol - released November 17, 2026, successor to GPT-6.1 Sol at the same
// Standard pricing ($2/M input, $10/M output, $0.10/M cache reads). Like Astra, it
// cannot disable reasoning. Keep the durable gpt/sol aliases on the latest Sol tier
// without moving users to the more expensive Astra tier. GPT-6.1 Sol stays usable as
// the custom model string `openai:gpt-6.1-sol`. OpenAI GPT models older than GPT-6
// (and GPT-6 Sol) are no longer supported: Pro reasoning mode on the GPT-6 tiers
// supersedes the separate gpt-5.5-pro model id.
// https://developers.openai.com/api/docs/models/gpt-6.2-sol
GPT: {
provider: "openai",
providerModelId: "gpt-6.1-sol",
providerModelId: "gpt-6.2-sol",
aliases: ["gpt", "sol"],
warm: true,
// GPT-6's tokenizer is not published upstream; reuse gpt-5 for approximate counting.
tokenizerOverride: "openai/gpt-5",
// OpenAI requires Daybreak Red approval for reduced refusals, yet the model
// rejects daybreak_red.
cyberAccessProgram: "daybreak_blue",
},
// GPT-6 Luna - the latest cost-efficient tier, released September 22, 2026.
GPT_6_LUNA: {
Expand All @@ -152,16 +140,18 @@ const MODEL_DEFINITIONS = {
aliases: ["luna"],
tokenizerOverride: "openai/gpt-5",
},
// Astra stays a separate frontier-tier choice rather than taking over `gpt`.
// Not warmed: its tokenizer override is already warmed via GPT.
// GPT-6.2 Astra - released November 17, 2026, successor to GPT-6 Astra at lower
// Standard pricing ($8/M input, $40/M output). Astra stays a separate frontier-tier
// choice rather than taking over `gpt`. The version alias follows the version, as
// `grok-4.7` did: GPT-6 Astra stays usable as the custom model string
// `openai:gpt-6-astra`. Not warmed: its tokenizer override is already warmed via GPT.
// https://developers.openai.com/api/docs/models/gpt-6.2-astra
GPT_6_ASTRA: {
provider: "openai",
providerModelId: "gpt-6-astra",
aliases: ["astra", "gpt-6-astra"],
providerModelId: "gpt-6.2-astra",
aliases: ["astra", "gpt-6.2-astra"],
// GPT-6 tokenizer not published upstream; reuse gpt-5 for approximate counting.
tokenizerOverride: "openai/gpt-5",
// Same Daybreak constraint as GPT (see its cyberAccessProgram).
cyberAccessProgram: "daybreak_blue",
},
// Gemini 3.1 Pro supersedes Gemini 3 Pro; keep bare aliases pointed at the latest Pro tier.
GEMINI_31_PRO: {
Expand Down Expand Up @@ -313,6 +303,8 @@ const LEGACY_TOKENIZER_MODEL_OVERRIDES: Record<string, string> = {
"anthropic:claude-sonnet-5": "anthropic/claude-sonnet-4.5",
"anthropic:claude-haiku-4-5": "anthropic/claude-3.5-haiku",
"anthropic:claude-opus-4-8": "anthropic/claude-opus-4.5",
"openai:gpt-6.1-sol": "openai/gpt-5",
"openai:gpt-6-astra": "openai/gpt-5",
"openai:daybreak-blue-latest": "openai/gpt-5",
"openai:daybreak-red-latest": "openai/gpt-5",
};
Expand Down
Loading
Loading