Here are all of the features introduced in each release. Have a feature request or feedback? Email us: help@double.bot

Note that not all of the changes here are reflected as extension versions, some changes are only on our backend, website, and other non-extension related services.

We’re working on something big… 👀

​
Week of October 21st, 2024

  • Added support for the latest Sonnet 3.5 model (2024-10-22)

​
Week of September 30th, 2024

Features:

  • You can now stop a CodeSnap mid completion by clicking on the CodeSnap button again.

Improvements:

  • Chat now supports and renders LaTex
  • Buttons in the Chat sidebar, including CodeSnap, copy to clipboard, and warp text are now easier to click

​
Week of September 23rd, 2024

Features:

  • CodeSnap v2:
    • CodeSnap will now automatically identify the target file for applying code instead of using the currently active editor.
    • Reliability has been improved to eliminate occasional unwanted commentary and comments in the merged code, like cases that started with “Here’s your updated code: \n ```python\n”
  • Added latest checkpoint of GPT4o (2024-09-03), which is above Sonnet 3.5 on Chatbot Arena Leaderboard for coding!
  • You can now subscribe to Double with Crypto!

Improvements:

  • More documentation in the tutorial on auto-import for autocomplete in scenarios where the user does not have an LSP running.
  • UI Polish:
    • Fix theming inconsistencies for light themes for ghost buttons and conversation titles in the dropdown list
    • Unescape markdown in conversation list title

Bug Fixes:

  • Fix theming inconsistentcies for buttons on top of the chat, and the links in the conversation list dropdown.

​
Week of September 16th, 2024

Bug Fixes:

  • Fixed an issue where combination of autosave + format on save inteferes with Inline Edit and CodeSnap.

​
Week of September 9th, 2024

Features:

  • Added OpenAI’s new o1-preview model to our selection of supported models.

  • Added support for Custom Instructions in chat, allowing users to set personalized guidelines for AI responses. Users can now specify preferences for tone, format, or specific content requirements that persist across conversations.

  • You can now switch models within the chat UI

Bug Fixes:

  • Updated frontend model name from DeepSeek Coder V2 -> DeepSeek V2.5 to accurately reflect the upgrade in backend last week.
  • Fixed an issue where Inline Edit removes lines following the edit region.
  • Fixed issues with optimistic updates of new messages:
    • Fix draft msg not being repopulated after a request fails
    • Fix optimistic messages not being removed after a request fails

​
Week of September 2nd, 2024

Features:

  • DeepSeek Coder V2 is upgraded to DeepSeek V2.5. DeepSeek V2.5 merges DeepSeek’s Chat and Coder models, improving alignment with human preferences and instruction following, as well as improved code generation capabilities compared to the original Coder model.

DeepSeek V2.5 Improvements

MetricDeepSeek-Coder-V2-0724DeepSeek-V2.5
AlpacaEval 2.044.550.5
ArenaHard66.376.2
AlignBench7.918.04
MT-Bench8.919.02
HumanEval python87.289
HumanEval Multi74.873.8
LiveCodeBench(01-09)39.741.8
Aider72.972.2
SWE-verified1916.8
DS-FIM-Eval73.278.3
DS-Arena-Code49.563.1

Improvements:

  • Improved responsiveness of the UI when working with long chats

​
Week of August 26th, 2024

Improvements:

  • Added support for autocomplete for .hbs files.
  • Improved performance and responsiveness when working with large files and large code contexts in chat.

Bug Fixes:

  • Fixed a bug where the last message occasionally disappears or flashes
  • Fixed a bug where a chat becomes bricked when the AI reply is empty after being cancelled
  • Fixed a bug where syntax highlighting will flash when applying code through CodeSnap or Inline Edit
  • Fixed a bug where codeblocks in chat sometimes does not match user’s theme

​
Week of August 19th, 2024

Features:

  • Launched CodeSnap, a new way of applying changes from chat to your code with 1 click. No more having to copy-paste from the chat window and figuring out what goes where. Now available to all users.

​
Week of August 5th, 2024

Features:

  • Added OpenAi’s new gpt-4o-2024-08-06 to our selection of supported models. You can read more about the model here. Launch tweet here.
  • Added a button to the Chat textbox to stop generations on command. See this tweet for a video demo.

​
Week of July 29th, 2024

Improvements:

  • Improved the formatting of titles, subtitles, bullet points, and spacing in the Chat modal. See this tweet for a video example of the improvements!

​
Week of July 22nd, 2024

Features:

  • Added Mistral Large 2, and DeepSeek Coder V2 (7/24) to our selection of supported models.
  • Added Llama 3.1 405B, the first frontier-level open-source AI model with 128k context length, to our selection of supported models. Also added Llama 3.1 70B and LLama 3.1 8B.

​
Week of July 15th, 2024

Features:

  • Fixed a bug in Jupyter notebooks where copy-paste from Chat would not work.
  • Fixed various other bugs that would occasionally cause issues connecting to our servers.
  • We have released a new custom Autocomplete AI model and are rolling it out to all users over the next week.

​
Week of July 8th, 2024

Improvements:

  • Disabled sticky autoscroll for Chat based on user feedback.
  • We now stream the code suggestions made by Inline Edits (preview) so that results are shown a lot faster.
  • Changed the default model for Inline Edits (preview) to Claude 3.5 Sonnet for faster and higher quality code generation.
  • Increased the maximum per message token count for Chat from 4,000 to 20,000 tokens.

​
Week of June 24th, 2024

Improvements:

  • Overall platform reliability improvements, and streamlining and bug fixing to continue improving the new Inline Edits (preview) feature.

​
Week of June 17th, 2024

Features:

​
Week of June 10th, 2024

Bug Fixes:

  • Reliability improvements for Inline Edits (beta). Fixed unnecessary empty lines always preceding the edit, and subsequent lines disappearing after accepting an edit.

​
Week of June 3rd, 2024

Improvements:

  • Improved Autocomplete caching. Cache hit rate should now be higher and respond faster. Overall Autocomplete should feel faster.

​
Week of May 27th, 2024

Features:

  • Launched Inline Edits for beta testing. This feature allows you to highlight any code, press Option+O, write your instructions, and see the output in diff style.

​
Week of May 13th, 2024

Features:

​
Week of April 22nd, 2024

Improvements:

  • Removed the need for SMS verification for new users creating an account.

​
Week of April 15th, 2024

Features:

​
Week of April 8th, 2024

Improvements:

  • UI improvements on how code snippets are shown in the Chat history.
  • Added the new and improved GPT-4 Turbo (2024-04-09) model from OpenAI, benchmarks indicate that this is OpenAI’s most capable coding model.
  • Started this changelog! 👋

Was this page helpful?

ModelsDouble Blog