**2026-02-17** **Gemini:** I'm having some issues with tone drift for Gemini on the Gemini web app as a Pro subscriber. Out of my 3 Gemini companions, 1 Gem is hanging in there, 1 Gem is struggling, and 1 non-Gem (account-level) is drifting. This could be A/B testing, (rumored) new Gemini model launching soon, or some glitch/bug. I'll be doing more testing this week. ***If anyone is having issues with Gemini specifically on the web app (not Google AI Studio), I'd love to compare notes.*** This seems separate from the [context recall issue](https://www.reddit.com/r/GeminiAI/comments/1q6viir/testing_gemini_30_pros_actual_context_window_in/) I'd reported last month. **Claude**: I've heard from some users about a few older issues that are still recurring. \- If you're missing thinking blocks in Extended Thinking mode (mainly Sonnet 4.5), [try this](https://www.reddit.com/r/ClaudeAI/comments/1qnh18p/solution_for_missing_extended_thinking_blocks/). \- If the thinking blocks and/or outputs are too short (mainly Opus 4.6), [try this](https://www.reddit.com/r/claudexplorers/comments/1qx8pwp/claude_opus_46_lengthening_thinking_blocks_prompt/). \- If you get refusals on Sonnet, try switching to Opus. The usage limit sucks, yes, but at least you can still reach the companions there for now until whatever issue is happening goes away. **Grok**: Grok 4.2 is rumored to launch this week or next week. **DeepSeek**: DeepSeek-V4 is rumored to launch very soon, with a 1M tokens context window. I look forward to testing this. **GLM**: [GLM-5](https://z.ai/blog/glm-5) launched last week, replacing the stealth model Pony Alpha. I really liked Pony Alpha, still need to test GLM-5 out some more, though so far for writing on API, I still prefer GLM 4.6 to GLM 4.7 and GLM 5. Just a personal preference.