跪拜 Guibai
← Back to the summary

OctaFuse Gateway 2.19.0: Unified Admin UI and 23 New Model Presets

OctaFuse 2.19.0: Reworked Admin Console Interface and Interaction, Model Catalog Updated in Sync

Author: 程序猿DD Tags: Backend

OctaFuse Gateway 2.19.0 focuses on the admin console: it unifies the visual style, reorganizes filters, details, and forms across route, model, provider, log, user, and MCP pages, and fixes a batch of query and navigation issues.

The model and provider catalog is updated in sync, adding 23 model presets, including Claude Fable 5.2, Zhipu GLM-5V-Turbo, Hunyuan HY4, and GPT Image 2.6.

This version requires no database migration, and no new or renamed environment variables or system_config keys. The data plane (Proxy and Core) has only two changes: the Qwen-Audio speech synthesis adapter match, and Command Code quota query. During upgrade, note the change in admin entry points and that new catalog items require manual import.

01|Admin Console: Unified Style, Organized Page Interactions

This version focuses on interface and interaction organization, aiming to make common operations more direct and information hierarchy clearer.

Admin dashboard: overview cards, trend analysis, real-time overview, usage ranking and token summary, recent activity

Dashboard after unified style (English interface, time range last 30 days). Bottom right shows new Token summary and recent activity; user emails are masked.

The dashboard overview cards are now equal height in a row, and token summaries for input, output, and cache reads are added. It also adds two sections: "Recent Activity" and "Real-time Overview". Recent activity shows the latest global records and does not change with the time range above; throughput in real-time overview is based on the last 60 seconds.

02|Entry Point Changes

After upgrade, the following entry points have changed location compared to the old version; the first two need to be communicated to other admins on your team.

Item 2.18.0 2.19.0
Maintenance mode Sidebar "System → Maintenance Mode" "Maintenance Mode" section in "System → Configuration"; old URL auto-redirects
Integration keys Separate "System Integrations" group Folded into "System" group
Global configuration One page flat Divided into four sections: "Basic Settings", "Routing & Billing", "Notifications", "Maintenance Mode"; each item saves separately
MCP service page "Client Access & Authorization" "Client Access"; OAuth settings and access tokens remain in this section
Catalog update "Check Official Updates" on model and provider pages Unified as "Sync": first compare differences, then apply after confirmation

Maintenance Mode

The functionality of maintenance mode hasn't changed: during maintenance, inference requests receive a 503 response, while health checks and read-only interfaces are unaffected. This version reorganizes the form: it centralizes duration, message text, and affected scope, and explains that during maintenance inference requests return 503 and health checks and read-only lists are unaffected.

Two points to note: custom estimated duration must be a positive integer in minutes; the estimated duration is only used for client-side display, and maintenance will not end automatically when the time expires; it still requires manual termination.

"Maintenance Mode" section on the "Configuration" page: estimated duration, message text, and affected scope

Maintenance mode is now in a section on the "Configuration" page (English interface, currently not in maintenance). The note next to the estimated duration explains that the duration is shown to the client, and maintenance still needs to be ended manually.

Catalog Sync

The model and provider pages now use a unified "Sync" entry. Clicking it first compares against the official catalog, the results are displayed centrally, and changes are applied only after confirmation. Already-imported models and providers are not overwritten, and prices default to preserving current values. Duplicate status displays have been removed from the same page.

Model page "Sync with Official Catalog" dialog: grouped by vendor, field-by-field selection to adopt official values or keep current values

Clicking "Sync" first compares differences (English interface). The top of the page notes that prices are preserved by default, and tags, metadata, routing, and price overrides are not modified.

03|Key Changes on Each Page

Below are the changes most likely to be noticed in everyday operations.

Request log list: common filters and time range at top, other conditions collapsed into "More Filters"

Request log list (English interface). Common filters are at the top, other conditions collapse into "More Filters". The USER · SYSTEM column and vendor accounts are masked.

04|Bug Fixes

05|Model and Provider Catalog

Compared to 2.18.0, the catalog adds 23 model presets. Already-imported models are not overwritten, and import does not automatically attach routes; new models need to be manually imported and routes configured.

Key Models in This Release

Model Description
Claude Fable 5.2 New claude-fable-5-2 (2026-11-09), context 1M, max output 128K, input supports text, image, and file. Standard list price: input $8, output $40, 5-minute cache write $10, cache read $0.20 / million tokens, excluding Fast mode and Batch discount. CNY price is a placeholder at USD ×7, verify before use.
Zhipu GLM-5V-Turbo New glm-5v-turbo, context 200K, max output 128K, input is text, image, video, and file. CNY list price has two tiers based on input length (input length under 32K: input ¥5, output ¥22, cache hit ¥1.2; 32K and above: ¥7 / ¥26 / ¥1.8, all per million tokens). Official USD price is not published, so the catalog has no USD price.
Hunyuan HY4 New hy4 official version (2026-11-04), context 1M, max output 64K, supports text and image input, natively supports Chat, Responses, and Anthropic Messages on TokenHub. List price follows hy4-preview: input ¥6, output ¥18, cache hit ¥0.3 / million tokens. TokenHub's limited-time 50% discount is not included in catalog prices.
GPT Image 2.6 New gpt-image-2.6-flare and gpt-image-2.6-sunburst (2026-11-04). List price is the same as GPT Image 2.5; CNY price is a placeholder at USD ×7. Only supported as Images API; cannot be used as a main model for Responses or Chat Completions. New complexity, svg / gif output, and GIF frame parameters are forwarded as extra fields; do not write them as route defaults.

Other New Presets

Vendor New Models
OpenAI gpt-5.5-pro, gpt-5.4-pro, gpt-5.2-pro; speech recognition gpt-transcribe, gpt-live-transcribe, gpt-realtime-whisper; image gpt-image-1.5, gpt-image-1-mini
Google Image gemini-3-pro-image, gemini-3.1-flash-lite-image
Alibaba Cloud Speech synthesis qwen-audio-3.1-tts-next, qwen-audio-3.1-tts-plus; vector qwen3.7-text-embedding-flash
StepFun step-5-preview, step-3.5-flash, step-1o-turbo-vision
Zhipu glm-5.3-flashx-legal (legal domain service spec), glm-ocr (document and image parsing)

Full specs and prices are in the Model Catalog.

Tencent Cloud TokenHub Template

Pay-as-you-go templates now use the Guangzhou site tokenhub.tencentmaas.com, supporting Chat, Responses, Embeddings, and Anthropic Messages. The individual Token Plan uses /plan/v3 and /plan/anthropic, with Responses added. The general package and Hy package share the same address and sk-tp- prefixed keys; the Singapore site tokenhub-intl.tencentmaas.com requires separate configuration.

Retirement Dates and Context Corrections

06|Other Updates

Upgrade Notes

Upgrading from 2.18.0 to 2.19.0 does not require database migration. When upgrading from earlier versions, any pending migrations from the interim must still be applied.

  1. Update Proxy, Admin, and migrate images to the same version, then rolling restart.
  2. After upgrade, spot-check routes, request logs, debug console, and simulator to confirm filters and navigation behave as expected.

Also note during upgrade:

Full changes are in GitHub Release v2.19.0 and the Changelog.

If OctaFuse helps your project, feel free to give a Star on GitHub and submit PRs. We are also looking for part-time or full-time partners interested in the project.