LLMTranslate 0.4.0
Bug Fixes
- Fixed a crash during translation when a prompt referenced an
unresolved placeholder (e.g.
{items_text} in item-by-item
mode) or when survey text contained literal curly braces. Prompt
interpolation now leaves unknown placeholders untouched instead of
aborting the run.
- Fixed GPT-5 / o-series (“reasoning”) models: requests previously
failed and silently fell back to a chat model. The ‘responses’ endpoint
payload and the output extraction are corrected, and the output token
budget was raised so large batch responses are no longer truncated.
New Features and
Improvements
- API calls now automatically retry transient failures (HTTP 429 rate
limits and 5xx server errors) with exponential backoff, and surface the
provider’s error message (e.g. quota details) instead of a bare status
code.
- Custom (user-typed) model names that are not in the built-in list
now work end to end: the provider is inferred from the model name, so
newly released OpenAI, Gemini, and Anthropic models can be used without
an app update.
- Updated the built-in model list to current models (e.g. Claude Opus
4.8, Sonnet 4.6, Haiku 4.5). Anthropic models that do not accept a
temperature parameter are handled automatically.
- Model selection fields are now freely editable text boxes with a
suggestion dropdown, so a custom model name can be typed or edited
directly without first clearing the current selection.
LLMTranslate 0.3.0
Major Features
- Batch Translation Mode: New translation mode that
sends all items in a single LLM call per stage (forward/back/recon) for
faster processing and better context-aware translations
- Multi-Sheet Excel Support: Can now select and
translate individual sheets or all sheets at once from Excel files with
multiple sheets
- Custom Model Input: Model selection fields now
accept custom model names typed by users, allowing use of newly released
models without app updates
- Empty Row Handling: Automatically filters and skips
empty rows during translation while maintaining correct row
alignment
- Sheet-Specific Status: Status notifications now
indicate which sheet is being processed during multi-sheet
translations
UI/UX Improvements
- Batch Translation tab now appears before Item-by-item Translation as
the recommended default
- Added sheet selector dropdown for viewing results from different
sheets after multi-sheet translation
- Added “Translation Mode” column to Model Selection Log indicating
whether Batch or Item-by-item mode was used
- Added comprehensive Excel file preparation guide in Help tab
- Model selection dropdowns now support typing custom model names with
autocomplete
- Custom models trigger informative warnings instead of blocking
translation
- Progress text shows current sheet name during multi-sheet
processing
- Multi-sheet downloads include all translated sheets plus Model
Selection and Prompt logs
Translation Quality
Improvements
- Batch mode prompts emphasize context and terminology consistency
across items
- Fixed item number parsing in batch responses to prevent row
misalignment
- Parser now uses
item_number field from LLM responses
for accurate mapping
- Improved error handling and logging for batch translation
operations
Bug Fixes
- Fixed preview table error when displaying multi-sheet translation
results
- Fixed row mapping issue where translations were offset by one
row
- Fixed tryCatch structure in batch translation for proper error
handling
- Corrected indentation and control flow in multi-sheet processing
loop
LLMTranslate 0.2.0
Major Features
- Added support for Anthropic Claude models (Sonnet 4.5, Haiku 4.5,
Opus 4.1, and others)
- Updated Google Gemini models to latest versions (Gemini 2.5 Pro, 2.5
Flash, 2.0 Flash)
- Added Model Selection Log sheet to Excel output
- Added Prompt Log sheet to Excel output for full reproducibility
UI/UX Improvements
- Added visual progress bar showing item X of Y with percentage
- Added “Test Connection” buttons for all API providers with
success/error feedback
- Added resume functionality - can continue after stopping
mid-batch
- Added warning notification for large batches (500+ items)
- Added real-time preview of first completed translation
- Added “Reset” button to clear all data and start fresh
- Added comprehensive “API Keys Guide” tab with step-by-step
instructions for obtaining keys
- Improved Help tab with updated instructions
Bug Fixes
- Fixed Unicode display issues in API status messages
- Fixed deprecated Gemini model names (updated fallback logic)
LLMTranslate 0.1.3
LLMTranslate 0.1.2
LLMTranslate 0.1.1
- Address CRAN feedback: quotes around software names, expanded
acronyms, added references, in Rd, minimal test.
LLMTranslate 0.1.0
(2025-07-24)