You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* Include reasoning messages in cloud task histories so you can review complete thinking after runs ([#8401](https://github.com/RooCodeInc/Roo-Code/pull/8401))
20
+
* Add structured data to the homepage to improve SEO and discoverability ([#8427](https://github.com/RooCodeInc/Roo-Code/pull/8427))
21
+
22
+
## Bug Fixes
23
+
24
+
* “Reset and Continue” now truly resets cost tracking; subsequent requests only count new costs (thanks alecoot!) ([#6890](https://github.com/RooCodeInc/Roo-Code/pull/6890))
25
+
* Prompts settings: Save button activates on first change and preserves empty strings (thanks beccare!) ([#8267](https://github.com/RooCodeInc/Roo-Code/pull/8267))
26
+
* Remove overeager “unsaved changes” dialog in Settings when nothing changed ([#8410](https://github.com/RooCodeInc/Roo-Code/pull/8410))
27
+
* Show the Send button when only images are attached ([#8423](https://github.com/RooCodeInc/Roo-Code/pull/8423))
28
+
29
+
## Provider Updates
30
+
31
+
* Chutes: Add new models for speed, longer context, and multimodal reasoning (DeepSeek V3.1 Terminus, V3.1 turbo, V3.2-Exp; GLM‑4.6‑FP8; Qwen3‑VL‑235B‑A22B‑Thinking) (thanks mohammad154!) ([#8467](https://github.com/RooCodeInc/Roo-Code/pull/8467))
* Bedrock: Treat Claude Sonnet 4 as 1M context to reduce truncation and mis-selection ([#8421](https://github.com/RooCodeInc/Roo-Code/pull/8421))
34
+
* Roo Code Cloud: Deprecate “Grok 4 Fast”; show warning if selected and hide when not selected ([#8481](https://github.com/RooCodeInc/Roo-Code/pull/8481))
35
+
36
+
## Misc Improvements
37
+
38
+
* Add internal UsageStats schema and type to enable future dashboards and clearer usage insights ([#8441](https://github.com/RooCodeInc/Roo-Code/pull/8441))
39
+
* Security: Update Vite dev server to 6.3.6 to address path traversal (GHSA-g4jq-h2w9-997c) (thanks app/renovate!) ([#7838](https://github.com/RooCodeInc/Roo-Code/pull/7838))
40
+
* Dependency: Update glob to 11.0.3 for stability (thanks app/renovate!) ([#7767](https://github.com/RooCodeInc/Roo-Code/pull/7767))
Copy file name to clipboardExpand all lines: docs/update-notes/v3.28.mdx
+9Lines changed: 9 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -98,6 +98,7 @@ This gives teams a higher-capacity OpenAI option without extra configuration.[#8
98
98
***Cleaner prompts (no “thinking” tags)**: Removes `<thinking>` tags for cleaner output, fewer tokens, and better model compatibility; preserves the plain‑language rule to confirm tool success. ([#8319](https://github.com/RooCodeInc/Roo-Code/pull/8319))
99
99
***Simpler Requesty model refresh**: Removes an unnecessary warning when refreshing the Requesty models list so you can fetch new policies without leaving the screen. ([#7710](https://github.com/RooCodeInc/Roo-Code/pull/7710))
100
100
***Bedrock 1M Context Checkbox**: Adds checkbox to enable 1M context window support for Claude Sonnet 4 and 4.5 on Bedrock, providing more flexibility for large-scale projects ([#8384](https://github.com/RooCodeInc/Roo-Code/pull/8384))
101
+
* Include reasoning messages in cloud task histories so you can review complete thinking after runs ([#8401](https://github.com/RooCodeInc/Roo-Code/pull/8401))
101
102
102
103
## Bug Fixes
103
104
@@ -138,6 +139,10 @@ This gives teams a higher-capacity OpenAI option without extra configuration.[#8
138
139
139
140
***AWS Bedrock**: Removed `topP` parameter to avoid conflicts with extended thinking, improving reliability for streaming and non‑streaming responses (thanks ronyblum!) ([#8388](https://github.com/RooCodeInc/Roo-Code/pull/8388))
140
141
***Vertex AI**: Fixed Sonnet 4.5 model configuration to prevent 404s and align with the correct default model ID (thanks nickcatal!) ([#8391](https://github.com/RooCodeInc/Roo-Code/pull/8391))
142
+
* "Reset and Continue" now truly resets cost tracking; subsequent requests only count new message costs ([#6890](https://github.com/RooCodeInc/Roo-Code/pull/6890))
143
+
* Prompts settings: Save button activates on first change and preserves empty strings ([#8267](https://github.com/RooCodeInc/Roo-Code/pull/8267))
* Show the Send button when only images are attached ([#8423](https://github.com/RooCodeInc/Roo-Code/pull/8423))
141
146
142
147
## Provider Updates
143
148
@@ -152,6 +157,10 @@ This gives teams a higher-capacity OpenAI option without extra configuration.[#8
152
157
153
158
***Supernova 1M context**: Upgrades the default Supernova to roo/code-supernova-1-million (1,000,000‑token context). Existing settings auto‑migrate; expect fewer truncations on long tasks. ([#8330](https://github.com/RooCodeInc/Roo-Code/pull/8330))
154
159
***Z.ai GLM‑4.6**: Adds GLM‑4.6 model support with a 200k (204,800) context window and availability across both international and mainland API lines (thanks dmarkey!) ([#8408](https://github.com/RooCodeInc/Roo-Code/pull/8408))
160
+
***Chutes**: Add DeepSeek‑V3.1‑Terminus, DeepSeek‑V3.1‑turbo, DeepSeek‑V3.2‑Exp, GLM‑4.6‑FP8, and Qwen3‑VL‑235B‑A22B‑Thinking to the provider picker ([#8467](https://github.com/RooCodeInc/Roo-Code/pull/8467))
***Roo Code Cloud**: Deprecate “Grok 4 Fast”; show warning if selected and hide when not selected ([#8481](https://github.com/RooCodeInc/Roo-Code/pull/8481))
163
+
***Bedrock**: Claude Sonnet 4 1M‑context handling and compatibility improvements ([#8421](https://github.com/RooCodeInc/Roo-Code/pull/8421))
0 commit comments