117 ENTRIES · OLDEST 8 JANUARY 2026 · NEWEST 26 AUGUST 2026 · CHECKED NIGHTLY FROM 27 JULY 2026

What changed

This record holds changes to deployed AI systems. Entries use the company's own words or measurements made by us. Each entry says which.

Orientation

How to read this record

117 entriesReports can be submitted to research@nlnllabs.com.

Each entry gives the date, what changed in plain words, the company's own words, a link to their page, and a copy of that page saved by us with a fingerprint. If the company later edits or deletes the page, our copy still exists.

The public record

The record

showing 117 of 117

26 Aug Anthropic Sonnet 5 sessions now run 33,000 tokens longer before compacting

26 Aug Anthropic Sign-in enforcement now exits when it cannot read managed settings

26 Aug Anthropic Analytics now stay off from startup, not just after login

25 Aug Anthropic Computer use on macOS lost its Finder and Dock exemption

25 Aug Anthropic The sandbox prompt stopped telling Claude which hosts were allowed

25 Aug Anthropic Model and effort commands stopped waiting for the turn to end

25 Aug Anthropic An idle session now starts at most three check-ins per goal

22 Aug Anthropic Every "generally available" claim was rewritten across two years

21 Aug Anthropic The retry watchdog stopped waiting forever on a spend limit

21 Aug Anthropic Background check-ins were slowed down, then capped at three

20 Aug Anthropic Double-pressing Ctrl+L no longer clears the conversation

20 Aug Anthropic Listing MCP servers stopped connecting to the disabled ones

19 Aug Anthropic A file-hiding rule can no longer be beaten by renaming the file

18 Aug OpenAI The Model Spec's required shutdown timer became optional

17 Aug Anthropic A Windows path spelling was closed across every remaining entry point

17 Aug Anthropic The teammate model setting was removed from the config menu

14 Aug Anthropic Claude Code turned subagent forking on by default

14 Aug Anthropic Claude Code withdrew its todo tools from newer models

13 Aug OpenAI Persisted reasoning items could be decoded into plain text by a weaker model

13 Aug Anthropic Encrypted thinking blocks could be decoded by a weaker model instead of the server

13 Aug DeepSeek The deepseek-v4-pro alias now serves a newer build

13 Aug Anthropic Three model retirement notices were narrowed to apply to the Claude API only

13 Aug Anthropic Claude Code stopped auto-approving git and gh commands with dangerous flags

12 Aug Anthropic Claude Code let newer models overwrite files they had not read this session

10 Aug Anthropic Auto mode became Claude Code's default for new sessions on 14 August

10 Aug OpenRouter The auto router began picking models based on aggregate user spending

7 Aug Anthropic Fable 5's biology safety classifier was retrained to cut fallbacks by about 85 percent

6 Aug OpenAI GPT-5.6 Sol in ChatGPT began powering both instant and deeper reasoning replies

6 Aug Anthropic Claude Code closed permission-check gaps where commands hid part of themselves invisibly

6 Aug OpenAI The chat-latest alias now points to the model serving ChatGPT for Plus and Pro users

5 Aug OpenAI Selecting a cyber capable model in Codex began forcing safer permission defaults

4 Aug Anthropic Claude Code closed a Bash permission bypass hidden in zsh regex conditionals

4 Aug Anthropic Claude Code removed the Ultraplan research preview entirely

31 Jul DeepSeek The deepseek-v4-flash alias now serves a newer, retrained build

30 Jul OpenAI Priority Processing was renamed Fast mode and sped up by up to 2.5 times

24 Jul Anthropic Fast mode requests to Claude Opus 4.7 began returning an error instead of running

24 Jul Anthropic Disabling thinking on Claude Opus 5 began failing at the two highest effort levels

24 Jul OpenRouter The tool field on message endpoints widened from one shape to three options

19 Jul Anthropic Claude Code stopped auto-running the verify and code review skills at turn end

18 Jul OpenAI Codex lowered the posted context window for three models to 272,000 tokens

15 Jul OpenAI Custom instructions limit in ChatGPT raised from 1,500 to 5,000 characters

8 Jul OpenAI ChatGPT Voice switched to a new model without video or screen sharing support

6 Jul Anthropic Claude Code reverted /review to a fast single-pass check

6 Jul OpenAI ChatGPT's Instant rate-limit fallback switched to GPT-5.5 Instant Mini

6 Jul xAI Grok Voice launched 21 new voices in one release

1 Jul Anthropic Claude Code's Explore subagent switched from Haiku to the main session's model

1 Jul Anthropic Claude Fable 5 and Mythos 5 returned with a new exploit-blocking classifier

29 Jun Anthropic Claude Opus 4.6 dropped fast mode and matching requests began running at standard price

26 Jun Anthropic Claude API rate limits for Sonnet and Haiku rose to match Opus at every tier

26 Jun OpenAI ChatGPT dictation switched to a new speech-to-text model with a lower error rate

24 Jun Microsoft Copilot Free and Student plans dropped manual model choice for auto-only selection

24 Jun OpenAI GPT-5.5 Instant got better at carrying context across multiple turns

24 Jun OpenAI The chat-latest alias now points to a newer Instant model snapshot

22 Jun OpenAI ChatGPT's paste-to-attachment threshold rose from 5,000 to 10,000 characters

11 Jun Anthropic Claude Fable started telling users which model handled their query on fallback

8 Jun ElevenLabs ElevenAgents switched its default speech recognition provider to Scribe Realtime

4 Jun OpenAI ChatGPT memory switched from manual saved entries to automatic updates

1 Jun Google Gemini CLI's auto and flash aliases now resolve to Gemini 3.5 Flash

29 May OpenAI API prompt cache retention switched to a 24 hour default instead of in-memory only

28 May OpenAI GPT-5.5 Instant and Thinking lost canvas support for writing and coding

22 May Mistral Vibe CLI's default coding model switched from Devstral 2 to Mistral Medium 3.5

15 May OpenAI GPT-5.5 briefly performed worse for some users before a fix resolved it

15 May xAI Eight retired Grok API model aliases kept working but now serve different models

14 May Anthropic Claude Code's fast mode switched from Opus 4.6 to Opus 4.7 by default

13 May Microsoft VS Code fixed its context window indicator, which showed zero for BYOK models

10 May Perplexity Personal CFO's connectable account limit rose from 5 to 30

7 May Cursor Plan execution gained a parallel mode using async subagents

6 May Anthropic Claude Code's five hour rate limits were doubled for Pro, Max, Team and Enterprise plans

6 May Google The Interactions API replaced its outputs array response schema with a steps array

5 May Microsoft Viva Insights Power BI reports could now be published to anyone in the organization

5 May OpenAI GPT-5.5 Instant replaced GPT-5.3 Instant as ChatGPT's default model

24 Apr DeepSeek The deepseek-chat and deepseek-reasoner aliases now point to deepseek-v4-flash

22 Apr Anthropic Claude Code's default effort level for Opus 4.6 and Sonnet 4.6 rose from medium to high

18 Apr xAI Imagine began filtering moderated items without media out of its lists

16 Apr Anthropic Claude Code's default effort level was raised to xhigh for all plans

16 Apr Anthropic A verbosity cutting system prompt hurt Claude's coding quality alongside other changes

16 Apr Perplexity Claude Opus 4.7 became the default orchestrator model for Computer

11 Apr xAI Grok restored the bot challenge check on file upload

11 Apr xAI Imagine's NSFW content setting toggle was fixed

9 Apr OpenAI Codex usage on ChatGPT Plus shifted from long single day sessions to spread out sessions

8 Apr Microsoft Copilot's codebase search tool switched to a single auto managed semantic index

7 Apr Anthropic Claude Code's default effort level rose from medium to high for API and enterprise users

7 Apr OpenAI Codex stopped showing several legacy models in its picker for ChatGPT sign in users

4 Apr xAI Imagine's video extension start time was clamped to a minimum of two seconds

26 Mar Anthropic A bug made Claude clear its recent thinking every turn instead of after idle sessions

26 Mar Perplexity Computer began drafting output in Markdown by default instead of PDF

17 Mar OpenAI ChatGPT retired its Nerdy personality preset and moved users to the default

16 Mar OpenAI GPT-5.3 Instant reduced teaser style phrasing patterns in its replies

16 Mar OpenAI The gpt-5.3-chat-latest alias now points to whatever model ChatGPT currently uses

13 Mar Anthropic The 1M token context window became generally available for Opus 4.6 and Sonnet 4.6

13 Mar OpenAI GPT-5.4 got an image encoder fix for a bug in input image processing

12 Mar Anthropic Claude Code's default Opus model on Bedrock, Vertex and Foundry moved from 4.1 to 4.6

9 Mar Google The gemini-3-pro-preview alias now points to a different model

5 Mar Anthropic Claude Code moved Sonnet 4.5 users on Pro, Max and Team Premium plans to Sonnet 4.6

4 Mar Anthropic Claude Code removed Opus 4 and 4.1, moving pinned users to Opus 4.6

4 Mar Anthropic Claude Code's default effort level was reverted from medium back to high

3 Mar OpenAI GPT-5.3 Instant began avoiding dead ends, caveats and overly declarative phrasing

24 Feb Anthropic Claude Code's Bash tool began skipping the login shell by default

20 Feb OpenAI ChatGPT's Thinking mode context window grew from 196k to 256k tokens

19 Feb Anthropic Claude Code moved the Max plan's 1M context from Sonnet 4.5 to Sonnet 4.6

17 Feb Anthropic Sonnet 4.6's BrowseComp scores were revised down after a cheating detection fix

10 Feb Anthropic Opus 4.6's benchmark scores were revised down after a cheating detection pipeline update

5 Feb Perplexity Deep Research began running on Opus 4.6 instead of Opus 4.5

5 Feb Perplexity The enhanced memory engine's recall rate rose from 77% to 95%

3 Feb Anthropic Claude Code's /model command began running immediately instead of queuing

3 Feb OpenAI GPT-5.2 Thinking's Standard mode thinking time was shortened again

2 Feb OpenAI Codex CLI restored its default personality to Pragmatic

1 Feb xAI Grok-4-fast models' context limitation above 120k tokens was mitigated

29 Jan Anthropic Structured outputs became generally available for Claude Sonnet, Opus and Haiku 4.5

27 Jan Google Gemini 3 became the default model behind AI Overviews

22 Jan OpenAI GPT-5.2 Instant's default personality began adapting its tone more conversationally

21 Jan Google The gemini-pro-latest and gemini-flash-latest aliases now point to Gemini 3

13 Jan OpenAI The sora-2 alias now points to a newer dated snapshot

10 Jan OpenAI GPT-5.2 Thinking's accidentally lowered Extended thinking time was fixed

9 Jan OpenAI gpt-image-1.5 stopped forcing high fidelity on edits set to low fidelity

9 Jan OpenRouter The Auto Router's gateway p99 latency improved by roughly 70%

8 Jan Cursor Hooks in the CLI began executing about 10 times faster

This record does not claim to be complete. It holds what we found and could prove. From 27 July 2026 the check runs every night.

How the Ledger works

entries.json

research@nlnllabs.com