AI Provider for MiniMax -- WordPress 插件

开发者	mralaminahamed
更新时间	2026年9月12日 06:02
PHP版本:	7.4 及以上
WordPress版本:	7.1
版权:	GPL-2.0-or-later
版权网址:	版权信息

This plugin is an AI connector for MiniMax. It registers MiniMax with the WordPress AI Client, so every AI-enabled plugin on your site can generate text and images through MiniMax — text generation, chat, image understanding on MiniMax-M3, and image generation with image-01. This plugin is an independent, third-party integration and is not affiliated with, endorsed by, or sponsored by MiniMax. Why MiniMax? MiniMax is an AI company known for its high-performance, cost-effective language models. The MiniMax M2 and M3 series deliver frontier-level text generation quality — competitive with top models from OpenAI and Anthropic — while offering attractive pricing for high-volume WordPress use cases. MiniMax models are particularly strong at:

Long-context tasks — handling large documents, long conversations, and extended content generation
Instruction following — precise, structured output generation suitable for automated pipelines
Multilingual content — strong performance in English, Chinese, and other languages
Creative writing — rich, nuanced prose generation for blogs, marketing copy, and storytelling

Features

MiniMax M2 and M3 series models — including Highspeed variants optimised for faster responses
Automatic model discovery — live model list fetched from the MiniMax API and cached hourly; falls back to a hardcoded list when offline
Full parameter control — temperature, max tokens, top P, thinking mode, service tier, system instruction, and function declarations
Chat history — multi-turn conversations, not just single prompts
Image generation — image-01, with aspect ratio, orientation and multiple candidates
Text to speech — the speech-2.8, speech-2.6, speech-02 and speech-01 voices, with a choice of voice and of MP3, WAV, FLAC or Opus output
Settings page — configure default model and generation parameters without touching code
API key via Connectors — enter your key once in Settings > Connectors; all AI-enabled plugins share it automatically
Environment variable support — MINIMAX_API_KEY for server-level configuration, bypassing the database entirely
OpenAI-compatible API — uses the standard OpenAI API protocol for broad compatibility
Highspeed variants — M2.7 Highspeed, M2.5 Highspeed, and M2.1 Highspeed for latency-sensitive applications

What Can You Do With AI in WordPress? Once this provider is configured, any WordPress plugin or theme that integrates with the WordPress AI Client can use it:

Block editor (Gutenberg) — AI writing assistance, rephrasing, summarising, and expanding content
WooCommerce — generate product descriptions, SEO meta titles, customer review summaries, and category descriptions
SEO plugins — generate meta descriptions, focus keyphrases, and Open Graph content
Customer support — power AI chatbots and automate FAQ responses
Translation and localisation — translate and adapt content for English, Chinese, and other markets
Content marketing — generate blog post drafts, social media copy, and email newsletters
Image alt text — generate accessible alt attributes for media library images

Supported Models When an API key is configured, the live model list is fetched directly from the MiniMax API, so you always see the latest models. If the API is ever unavailable, a built-in fallback list keeps everything working — the MiniMax-M3 flagship plus the M2 series (M2.7, M2.5, M2.1, and M2), each with an optional Highspeed variant tuned for lower latency, alongside image-01 for images and eight speech-* voices for text to speech. See the full model catalogue for every model ID. Not sure which to pick? See the FAQ below. Requirements

PHP 7.4 or higher
WordPress 7.0 or higher
A MiniMax account and API key

How the WordPress AI Provider System Works WordPress 7.0 introduced a built-in AI Client SDK. Plugins and themes call a standard API (e.g. "generate text from this prompt") without knowing which AI provider is active. Provider plugins like this one register themselves with WordPress and handle the actual API calls. This means:

Install this plugin → MiniMax is registered as an AI provider
Enter your API key in Settings > Connectors
Every AI-enabled plugin on your site can now use MiniMax automatically

You can also install multiple provider plugins and switch between them from the Connectors screen — no re-configuration of individual plugins needed. Settings Go to Settings > MiniMax to configure:

Default Model — the model used when no explicit model is requested by a plugin
Temperature — controls output randomness (0.0–2.0, default 0.7)
Max Tokens — maximum tokens in the response (1–1,000,000, default 4096)
Top P — nucleus sampling threshold (0.0–1.0, default 1.0)
Thinking — whether MiniMax-M3 reasons before answering; disabling it is faster and cheaper (adaptive or disabled, default adaptive)
Service Tier — standard, or priority routing at 1.5x the cost (default standard)

For Developers This plugin follows the official WordPress AI Provider pattern and works with any plugin built on the WordPress AI Client SDK. It registers the minimax provider and supports the generation options MiniMax actually honours (temperature, max tokens, top P, system instruction, and function declarations). Presence and frequency penalties and stop sequences are deliberately not declared, because MiniMax documents them as ignored. The source code is on GitHub — bug reports and pull requests are welcome:

Repository: github.com/mralaminahamed/ai-provider-for-minimax
Report a bug or request a feature: open a new issue
Contribute code: submit a pull request

Developer documentation:

More from us Other free plugins by the same author, all on WordPress.org. Another provider for the same AI Client

AI Provider for OpenCode Zen - One API key, 61 models including GPT-5, Claude and Gemini 3, for the WordPress AI Client.

For any site

Swift Menu Duplicator - Duplicate menus in one click, snapshot revisions, export and import, WP-CLI and REST.
Author Profile Blocks - Author and team profiles as Gutenberg blocks — grid, carousel, list and single.

For a WooCommerce store

StoreSeeder - Realistic test data for WooCommerce and Fluent Cart — a whole shop from a recipe, with one-click cleanup.
StoreSheet - Sync WooCommerce products, orders and coupons to a Google spreadsheet, one way and in the background.

1.6.0 - 2026-09-12 Added

Text to speech. MiniMax's eight speech-* models are now offered as text-to-speech models, served from MiniMax's own /v1/t2a_v2 endpoint. Choose a voice, an output format (MP3, WAV, FLAC or Opus) and inline audio or an expiring URL. Speed, volume, pitch, emotion, sample rate, bitrate and language boost are reachable through custom options or the new minimax_convert_text_to_speech_params filter. Only the synchronous endpoint is implemented, so its 10,000-character limit is refused up front rather than reported by MiniMax as an invalid parameter.
A Test connection button on the settings page. The page used to say "MiniMax is connected" as soon as a key existed anywhere, without ever asking MiniMax — a revoked, mistyped or wrong-account key read as connected, and the first sign of trouble was a generation failing somewhere else. The check asks for the model list, which needs authentication and generates nothing, so it costs neither tokens nor money. A network failure is reported as "could not be checked", not as a bad key. The verdict is cached for five minutes and dropped whenever the key changes.
An uninstall handler. Deleting the plugin now removes the minimax_settings option and the minimax_models_cache transient, which it used to leave behind for good. Your API key is deliberately left alone: it belongs to the WordPress Connectors screen and, in the shared case, to every AI provider on the site. Changed
The plugin autoloads its own classes and ships no vendor/ directory. There was never a runtime dependency to justify Composer here.
The settings page no longer claims the provider is connected on the strength of a key existing. It says a key is configured, notes that this is not the same as one that works, and offers to check.
The default-model dropdown lists text models only. It is labelled "for text generation" but listed everything, so image-01 was already offered as a text model and the speech models would have added eight more.
Declared as tested against WordPress 7.1.
Directory listing copy: the title, tags and short description now say "connector" and name image generation, which is what people search for. Fixed
The chosen default model now decides which model is used. The setting has been stored since 1.0.0 and never read: picking one wrote a row to the database and changed no request that followed. Nothing is filtered, so a caller that names another model still gets it.
Generation requests are now addressed to MiniMax. The plugin returned a relative path where the AI Client expects an absolute URL, so every text and image request named no host and could not be sent.
A copy installed from git no longer activates and silently does nothing. It used to look for Composer's autoloader, find none, and return without registering a provider or saying why.
Refreshed the translation template, which still carried the old provider description.

1.5.0 - 2026-08-09 Added

Settings are now applied to requests. Temperature, max tokens and top_p had been stored since 1.0.0 and never read — nothing outside the settings class touched minimax_settings, so saving the form changed a database row and nothing else. A caller's own value still wins; the saved values fill in what was left unset.
Thinking mode — MiniMax-M3 can be told to answer without reasoning first, which is faster and cheaper. The M2 series always reasons and ignores the setting.
Service tier — opt in to priority routing, which MiniMax bills at 1.5x standard.
minimax_generate_text_params filter, for anything this plugin does not model.
Image input on MiniMax-M3. Vision has worked since the first release — the AI Client turns an image into an image_url content part — but was never declared, so image prompts were routed to other providers. The M2 series is text only and is declared as such.
Custom options. Arbitrary passthrough parameters, which the AI Client's base class already merged into the request body but which no caller could reach, because the option was not declared. All three official WordPress AI providers offer this.
Image generation. image-01 is now offered as an image model, served from MiniMax's own /v1/image_generation endpoint. Aspect ratio, orientation, number of candidates, and a choice between a URL and inline base64 are all supported. Reach seed, prompt_optimizer and subject_reference through the new minimax_generate_image_params filter.
Chat history — the models are now declared as supporting multi-turn conversations, not just single prompts. They always could (the OpenAI-compatible endpoint takes a messages array and the AI Client already sends one), but the capability was never declared and the AI Client routes on the declaration, so conversation requests were going to other providers. Every official WordPress AI provider declares this.

Changed

PHP namespace is now MiniMax\MiniMaxAiProvider\ (was AlAminAhamed\MiniMaxAiProvider\).
Class names dropped their MiniMax prefix, which the namespace already carries: MiniMaxProvider is Provider, MiniMaxSettings is Settings, MiniMaxModelMetadataDirectory is ModelMetadataDirectory, MiniMaxTextGenerationModel is TextGenerationModel, and MiniMaxProviderAvailability is ProviderAvailability.
Max tokens now accepts up to 1,000,000, matching MiniMax-M3's context window. The previous ceiling of 200,000 was the M2 limit.
Restructured the bootstrap. The plugin file is now an entry point — constants, autoloader, boot — and all wiring moved into an AI_Provider_For_MiniMax singleton in class-ai-provider-for-minimax.php, so every hook the plugin registers is visible in one file. Matches the layout used across this author's other plugins.
Added MINIMAX_VERSION, MINIMAX_URL and MINIMAX_PATH constants; only MINIMAX_PLUGIN_FILE existed before. Both renames are internal. No hook, option, setting or model id changes, and nothing a site has configured is affected — but any code referencing these classes directly needs updating.

Removed

Presence penalty and frequency penalty settings. MiniMax documents both as ignored on its OpenAI-compatible endpoint, so the controls could not affect anything. Saved values are left in the database untouched.
presencePenalty, frequencyPenalty and stopSequences are no longer declared as supported model options. The AI Client uses those declarations to decide which model can satisfy a request, so claiming them routed callers here and had their request quietly ignored. A caller that needs stop sequences is now correctly routed elsewhere.

1.4.0 - 2026-07-21 Added

Connection status on the settings page — shows whether a MiniMax API key is configured, with a link to the Connectors screen.
"Connectors" quick link on the Plugins screen, next to Settings. Changed
Synced the built-in fallback model list to the official MiniMax catalogue — added MiniMax-M3 and removed the retired MiniMax-M1 and MiniMax-Text-01 (8 models).
Default model is now MiniMax-M3, pre-selected out of the box instead of an empty choice. Fixed
Autoloader collision with WordPress core's bundled AI Client that could cause a TypeError on AI requests — the plugin no longer ships a conflicting copy of php-ai-client (WordPress core provides it at runtime).

1.3.1 - 2026-06-16 Fixed

API key stored via Settings > Connectors (connectors_ai_minimax_api_key) now correctly used for live model list fetching — previously fell back to the hardcoded model list even when the Connectors key was set.
Settings get_settings() now returns all 6 default fields when the saved option is corrupt or missing; previously returned only temperature and max_tokens.

1.3.0 - 2026-06-16 Added

Top P, Presence Penalty, and Frequency Penalty settings fields on the admin settings page.
Full SupportedOptions coverage: temperature, top P, presence penalty, frequency penalty, stop sequences, system instruction, function declarations, and max tokens. Changed
Extracted all admin HTML markup to templates/admin/ for cleaner separation of logic and presentation.
Renamed plugin class directory from src/ to includes/ per WordPress plugin conventions.
Removed AI Client SDK from Composer production dependencies — WordPress 7.0+ provides it natively at runtime.
Updated Requires at least to 7.0.

1.2.1 - 2026-05-01 Fixed

Connector showing as "Connected" before any API key is entered — provider availability now correctly checks for a configured API key.

1.2.0 - 2026-04-01 Added

Provider logo displayed on the WordPress Connectors page.
Expanded test suite from 33 to 44 tests covering all model IDs, highspeed variants, provider logo path, and settings edge cases. Fixed
False "no valid connector" warning on the AI admin page when API key is set via the Connectors page.

1.1.0 - 2026-03-01 Added

Updated fallback model list to include all current MiniMax text generation models: M2.7, M2.7 Highspeed, M2.5, M2.5 Highspeed, M2.1, M2.1 Highspeed, M2, M1, and Text-01.
Domain Path header field to plugin file. Changed
Improved plugin file header field ordering per WordPress.org standard.
Updated tested up to WordPress 7.0.

1.0.0 - 2026-01-01

Initial release.
MiniMax provider registration with WordPress AI Client.
Dynamic model discovery with transient caching and fallback list.
Settings page for default model configuration.
Support for MINIMAX_API_KEY environment variable.

AI Provider for MiniMax – AI Text & Image Generation Connector

标签

下载

详情介绍:

安装:

升级注意事项:

常见问题:

更新日志: