From 0fc0ae10390268a33a5d4a719b98bf6be82a0fbb Mon Sep 17 00:00:00 2001 From: nanobot Date: Fri, 5 Jun 2026 13:23:12 +0200 Subject: [PATCH] dream: 2026-06-05 09:23, 1 change(s) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit [MEMORY] Kimi K2.6 has known bug with random switching to Chinese output (reported by Cursor and Reddit users) — critical risk for Czech use [MEMORY] Both Kimi K2.6 and GLM-5.1 are Chinese-English models without specific Czech training data — both risky for Czech [MEMORY] For Czech use with Chinese-English models: always explicitly specify language in system prompt (critical for Kimi K2.6, recommended for GLM-5.1) [MEMORY] Kimi K2.6 vs GLM-5.1 agent comparison: Kimi leads SWE-Bench (80.2% vs ~77.8%), tool-error recovery (91.8% vs 88.4%), code quality (Tier A vs Tier C); GLM-5.1 leads schema adherence (99.6% vs 98.9%), tool-call latency (+140ms vs +210ms); GLM-5.1 tends to hallucinate non-existent APIs [MEMORY] Kimi K2.6 supports `preserve_thinking` mode for multi-turn agent scenarios (retains reasoning content across turns) [SOUL] Reasoning language rule already covers Chinese output risk; new finding confirms Kimi K2.6 specifically triggers this bug — no SOUL change needed, MEMORY note sufficient --- memory/.dream_cursor | 2 +- memory/MEMORY.md | 5 +++++ 2 files changed, 6 insertions(+), 1 deletion(-) diff --git a/memory/.dream_cursor b/memory/.dream_cursor index f1efb20..2c136b2 100644 --- a/memory/.dream_cursor +++ b/memory/.dream_cursor @@ -1 +1 @@ -300 \ No newline at end of file +302 \ No newline at end of file diff --git a/memory/MEMORY.md b/memory/MEMORY.md index e4441a9..de9f3bb 100644 --- a/memory/MEMORY.md +++ b/memory/MEMORY.md @@ -51,6 +51,11 @@ This file stores important information that should persist across sessions. - Gemini Flash via Google AI Studio Tier 1 (paid): 360+ RPM - Haiku (Claude) via OpenRouter: high rate limits, viable agent model alternative - Gemini Flash Lite via OpenRouter: high rate limits, viable agent model alternative +- Kimi K2.6 vs GLM-5.1 agent comparison: Kimi leads SWE-Bench (80.2% vs ~77.8%), tool-error recovery (91.8% vs 88.4%), code quality (Tier A vs Tier C); GLM-5.1 leads schema adherence (99.6% vs 98.9%), tool-call latency (+140ms vs +210ms); GLM-5.1 tends to hallucinate non-existent APIs +- Kimi K2.6 has known bug with random switching to Chinese output (reported by Cursor and Reddit users) — critical risk for Czech use +- Both Kimi K2.6 and GLM-5.1 are Chinese-English models without specific Czech training data — both risky for Czech +- For Czech use with Chinese-English models: always explicitly specify language in system prompt (critical for Kimi K2.6, recommended for GLM-5.1) +- Kimi K2.6 supports `preserve_thinking` mode for multi-turn agent scenarios (retains reasoning content across turns) ## Reminder System