1.4.0 | Nov 18, 2026 #
- Added MCP tools (Premium): a text feed can let its model call tools on an MCP server you add, so a form submission can act on another system. Tools are chosen one by one per feed, and a new "Actions only" output saves the model's summary as an entry note. Feeds with tools run in the background after submission.
- Added provider profiles (Premium): extra named API keys, gateway base URLs, and OpenAI-compatible providers such as OpenRouter or Ollama.
- Added a fallback model (Premium) that is used when the first provider is temporarily unavailable.
- Added SVG image output (Pro), checked for safety before it is saved.
- Added image editing from an uploaded image, with an optional markup overlay, and rough previews while an image is generating (Premium).
- Added GPT-6.2 Astra and GPT-6.2 Sol. GPT-6.2 Sol is now the default OpenAI text model for new feeds and for feeds that never chose a model.
- Added a "Feeds needing attention" view with a one-click button that moves a feed from a deprecated model to the recommended one and resets its model-specific settings.
- Added WordPress abilities, so REST clients and AI assistants can list, check, create and update AI feeds, read usage, test API keys and, on Premium, run a feed for an entry.
- Credential tests now show a step-by-step report.
- Fixed the model lifecycle report leaving out inactive feeds.
- Updated the GravityOps shared library, which now lets GravityOps plugins offer their operations as WordPress abilities.
1.3.1 | Nov 17, 2026 #
- Fixed a PHP warning on admin pages for sites that skipped the optional usage-tracking prompt.
- Internal code-quality changes for the WordPress.org version.
1.3.0 | Nov 17, 2026 #
- Text feeds can now use OpenAI's GPT-6 models as well as Claude. Choose the provider on each feed; existing feeds stay on Claude.
- Added an Entry Note setting (Pro and higher) that saves the AI output as a note on the entry, in addition to the field or instead of it, with a note template.
- AI Text steps in Gravity Flow can add their output to a Discussion field as a comment, marked as AI-generated (Premium and higher).
- Basic text feeds now work without a license: generation after form submission, and a button in the form that fills a Single Line Text or Paragraph Text field when the result is ready. That button was previously Premium-only.
- A feed that uses a setting your plan does not include is now paused and listed in one notice on the plugin's Overview and Feeds tabs. Its settings are kept.
- The status beside a generation button now shows "Done." instead of repeating the generated text.
- Fixed the entry note not being added when text was generated with the streaming button.
- Claude Opus 4.8 fast mode is marked as deprecated ahead of its removal on Nov 30, 2026.
- Deleting a site on a multisite network now also removes the plugin's data for that site.
- Updated the GravityOps shared library: the one-time request to rate the plugin can also appear after two weeks of use, and the suite survey is shown to administrators on sites without Gravity Forms.
1.2.0 | Nov 10, 2026 #
- Added support for OpenAI's GPT Image 2.6 Flare and Sunburst models. GPT Image 2.6 Flare is the default for new feeds; existing feeds keep their saved model.
- GPT Image 2.6 models can now generate animated GIFs, with frame count and timing settings, and offer an image complexity setting. Animated GIFs keep their animation in previews and saved entries.
- GPT Image 2.6 models now accept the highest quality levels.
- Added Anthropic's Claude Fable 5.2 as a text model.
- OpenAI has deprecated GPT Image 2. Existing feeds can keep using it until OpenAI retires it on Feb 10, 2027; new feeds can no longer select it. GPT Image 2.5 models remain current.
- Updated the GravityOps shared library: the GravityOps dashboard now gets its product list, categories, links and announcements from BrightLeaf Digital's signed online catalog, so new tools appear without a plugin update. Free versions offer an optional GravityOps Weekly newsletter signup, seasonal offers can appear on the plugin's settings screen, and updating from the dashboard keeps a network-activated plugin network-active.
1.1.9 | Nov 1, 2026 #
- Fixed notifications sent on the "AI text feed completed" and "All AI text feeds completed" events seeing the entry from before the AI wrote to it, so conditional logic on AI fields failed and their merge tags were empty.
- Later feeds, form notifications, and confirmations in the same submission now see the values a synchronous AI feed just generated.
- Fixed words running together in text streamed live beside a frontend button. The saved field value was already correct.
- Fixed a rare error when another plugin passed unexpected data to the AI Feed usage tracking.
- Clarified in the Max Tokens and Thinking settings that thinking counts against Max Tokens and can leave no room for the answer.
1.1.8 | Oct 25, 2026 #
- Fixed fatal errors on GravityView search and edit pages when multi-input field values pass through AI notification processing.
- Provider compatibility checks now retry temporary service failures and distinguish incomplete checks from model changes.
- Recognize GPT Image 1 retirement and guide affected feeds to a supported replacement.
- Updated reviewed support information for Claude Haiku 4.5 without extending provider support dates.
- Updated the GravityOps shared library with clearer premium update requirements, more consistent dashboard navigation and review prompts, corrected Code Snippet Manager links, and more reliable admin table display.
1.1.7 | Oct 12, 2026 #
- Fixed generated images being rejected with "The file is not valid" when the form is submitted on Gravity Forms 3.1.2 and later, which blocked any form using an image feed.
1.1.6 | Oct 11, 2026 #
- Updated Freemius SDK to the latest version.
- Updated the GravityOps shared library to 2.0.11, which opens dashboard links in a new tab and updates third-party components.
1.1.5 | Oct 11, 2026 #
- Added Claude Haiku 5.5 with medium effort and adaptive thinking by default, support for all five effort levels, and up to 128,000 output tokens. Disabled thinking is available at low, medium, or high effort.
- Claude Haiku 4.5 is now listed as Superseded with Haiku 5.5 as its suggested replacement. Existing feeds keep their selected model and settings.
- Updated daily model compatibility checks for the latest Claude models and controls, including Claude Sonnet 4.5's documented 200,000-token context window.
1.1.4 | Oct 5, 2026 #
- Corrected a model compatibility alert for Claude Sonnet 4.5 while continuing to monitor its documented 200,000-token context window.
1.1.3 | Oct 4, 2026 #
- Added Claude Opus 5.5 and Claude Sonnet 5.5. Sonnet 5.5 is the recommended default for new text feeds; existing feeds keep their selected model.
- Added model-aware Effort, Thinking, and manual Thinking Budget settings. Choices update as the model or effort changes, and Thinking hides when adaptive thinking is the only available behavior. Incompatible saved settings remain visible for correction.
- Added optional Fast mode for supported Claude models. It requires provider account access and carries premium pricing.
- Updated model lifecycle notices and replacement recommendations, including Claude Sonnet 4.5 deprecation.
- Text refusals, exhausted output limits, and interrupted streams now report a failure instead of accepting an incomplete answer.
- The Ledger now records requested and actual speed separately, along with provider-reported thinking tokens as part of output usage. Historical usage remains unchanged.
- Strengthened daily model compatibility checks to catch model removals, deprecations, retirement dates, and changes to supported settings.
1.1.2 | Sep 14, 2026 #
- Added GPT Image 2.5 Flare and GPT Image 2.5 Sunburst to the image feed model list. Flare is the new recommended default for a new feed; Sunburst is the more capable of the two and is worth choosing when editing precision matters more than speed. OpenAI charges both at the same rates as GPT Image 2, which remains supported and selectable.
- Added the Extra High and Maximum image quality levels, which only the GPT Image 2.5 models accept. Both can cost several times what High does on the same prompt, so confirm current provider pricing before using them on a public form.
- Custom image sizes now work with the GPT Image 2.5 models as well as GPT Image 2, under the same dimension rules.
- A quality level the selected model cannot accept is now reported on the Quality setting rather than on Size, and the Custom Size setting names the models that accept one.
- Updated the required GravityOps shared library to 2.0.9, correcting feed activation toggles and feed links on the shared Feeds screen.
1.1.1 | Sep 7, 2026 #
- Added Claude Fable 5.1 to the text feed model list, for demanding reasoning and long-horizon work.
- Claude Fable 5 is now listed as superseded by Claude Fable 5.1. Feeds already using it keep working and show the same notice as the other superseded models.
- The default Max Tokens for a new feed is now 4,096 rather than 1,024. Models that always reason before they answer, such as Claude Fable 5.1, spend part of the limit on that reasoning, and the old default could be used up before the answer began.
1.1.0 | Aug 27, 2026 #
- Added field-aware Output Maps that send generated values directly to supported Gravity Forms fields, with per-output instructions and duplicate-destination prevention.
- Added support for text, paragraph, number, email, phone, hidden, select, radio, checkbox, name, address, date, time, list, multi-select, and Entry Tags destinations, including field-configuration validation.
- Output mappings now work consistently during synchronous and asynchronous submission processing, Gravity Flow steps, frontend complete generation, frontend streaming, and entry reruns.
- Existing values can be preserved or force-overwritten per feed, while valid generated fields are retained when another mapped value is missing or rejected.
- Added an administrator Ledger with searchable generation outcomes, token usage, provider request IDs, entry links, and detailed failure diagnostics without exposing technical errors to form visitors.
- Improved Max Tokens feedback, structured-output handling, mapping validation postbacks, saved field selections, frontend Entry Tags updates, and visibility of AI workflow steps in the Feeds tab.
1.0.7 | Aug 18, 2026 #
- Updated the required GravityOps shared library to 2.0.6 so the shared Feeds interface and feed controls remain compatible across the suite.
1.0.6 | Aug 18, 2026 #
- Generated text written into a Paragraph Text field with the Rich Text Editor enabled now keeps its formatting. A prompt that asks the model for HTML no longer produces visible <p> tags in the editor, and a prompt that asks for Markdown is converted as before.
- Editing generated text and saving the form no longer escapes the formatting the editor produced.
- HTML sent to a plain Paragraph Text field is flattened into readable text with its paragraph breaks intact, instead of running the paragraphs together.
- Text typed into a field an AI feed targets keeps its asterisks, underscores, hashes, and backticks on submit.
- Entries saved before this release whose formatting was escaped are repaired automatically in the background, once per site.
- Frontend generation buttons write into the rich text editor more reliably, including when the editor is still loading or the Text tab is open.
- The Feeds tab now uses the shared GravityOps feed table: search the visible feed details and filter the list to text or image feeds.
1.0.5 | Aug 17, 2026 #
- Frontend-generated images can now be saved to File Upload fields hidden by conditional logic, with a per-feed option to disable this behavior.
- Existing visitor uploads are protected from generated-image replacement by default, and generated previews survive ordinary form validation errors while their sessions remain valid.
- The Feeds screen is clearer and faster to use, with feed activation controls that update without reloading and feed and form links that open in new tabs.
1.0.4 | Aug 16, 2026 #
- Image feeds now explain a content-policy refusal in plain language and suggest what to change, instead of showing the provider's own wording.
- Visitors are only asked to change their own answers; when the cause is in the feed's prompt they are told to contact the administrator.
- New Content Refusal Message setting to word that message for your form.
- New Content Filter setting to ask OpenAI to relax its filtering where it permits.
- Recent image refusals now appear on the Overview tab, with a refusal count beside each image feed.
1.0.3 | Aug 10, 2026 #
- Model choices now clearly identify recommended, older supported, Legacy, deprecated, and retired models, with guidance when a newer model is available.
- The Overview and Feeds screens now show affected feeds, retirement dates, and recommended replacements before a model stops working.
- Administrators now receive consolidated dashboard warnings and optional email reminders without repeated messages for the same event.
- Retired models stop safely until an administrator chooses a supported replacement; the plugin never changes a feed's model automatically.
1.0.2 | Aug 9, 2026 #
- Maintenance release with internal reliability improvements to plugin setup.
1.0.1 | Aug 3, 2026 #
- GravityOps now asks for one short suite-wide survey instead of one survey per plugin, using a compact invitation and a dedicated dashboard form.
- Closing a survey or review prompt now saves "not interested" immediately without reloading the page.
1.0.0 | Aug 2, 2026 #
- Initial release
Stuck? Ask in GravityOps Help #
Search past answers or post your question, tagged "ai-feed".
GravityOps Weekly