What's New
All updates and improvements to ThinkTower, version by version.
What's New in v0.93.10
•Robust model downloads — stall detection, automatic resume, and retry support
•Data card in chat — file badge visible in messages when working with data
•Fixed Excel loading on Intel Mac ("Unknown error" resolved)
ThinkTower v0.93.9
Downloads
•Apple Silicon (M1/M2/M3/M4): `ThinkTower-0.93.9-Mac-arm64.dmg`
•Intel Mac: `ThinkTower-0.93.9-Mac-x64.dmg`
•Windows: `ThinkTower-0.93.9-Windows-Setup.exe`
Wijzigingen
•Fix: startup timeout verhoogd voor Intel Macs (CPU-only modus)
•Fix: health checks stoppen nu correct na server failure
•Fix: verwijdering dode dev-only UI elementen
•Fix: LLM status tooltip werkt weer in alle states
•Fix: architectuur-specifieke builds (arm64/x64 gescheiden)
•Fix: native module verificatie in build pipeline
Changes
Model Downloads Fixed
•Fixed SHA256 hash mismatch for qwen3.5-27b (16-31GB machines)
•Fixed 404 download error for qwen3.5-35b (32-63GB machines) — file renamed by provider
•Replaced qwen3.5-122b MXFP4_MOE with UD-Q3_K_XL (64GB+ machines) — old file removed by provider
•All model downloads now pinned to specific git revisions to prevent future breakage
•Added multi-part GGUF download support for 122B model (3 parts, 53.1 GB)
Startup Reliability
•Window now always appears even if background services fail to initialize
•Error dialog shown if services fail (helps with bug reports)
Orphaned Model Cleanup
•App detects leftover model files from previous versions
•Asks user permission before deleting (frees disk space)
Build Improvements
•Pre-build check verifies all model files are available on HuggingFace
•mmproj (vision) downloads now use pinned revisions
Security
•Updated underscore dependency (GHSA-qpx9-hpmf-5gmw)
What's New
•Context length meter — shows token usage above chat input (green/orange/red bar)
•Trail panel — collapsed by default, clickable spacer to expand
Fixes
•Fix: app window not showing on some machines after upgrade (blank menu bar)
•Fix: Qwen3.5 template error with document/data context (system message conflict)
•Fix: remove CPU-only mode for 8GB machines (4B model has enough headroom)
•Fix: security update underscore 1.13.8 (GHSA-qpx9-hpmf-5gmw)
Downloads
| Platform | File |
|
v0.93.6
New
•Context length meter above chat input (shows token usage with color thresholds)
•Trail panel default collapsed + clickable spacer to expand
Fixed
•Qwen3.5 system message template error with documents, data, and schrijven mode
•Vision chat audit log crash (chat_vision constraint)
Improved
•GPU offloading re-enabled for 8GB machines (4B model leaves enough headroom)
•Markdown spacing (tighter headings, paragraphs, lists)
•Image message layout in chat (vertical stacking)
v0.93.5
•Vision support from 24GB+
•New model tiers: 4B replaces ministral-8b for 8-15GB machines
•Qwen3.5 vision/image support with mmproj
•Model selection bugfixes in settings modal
ThinkTower v0.93.4
Downloads
•Apple Silicon (M1/M2/M3): `ThinkTower-0.93.4-arm64.dmg`
•Intel Mac: `ThinkTower-0.93.4-x64.dmg`
Performance Fix
•Fixed: Fast startup on 8GB RAM machines
•Context size now automatically scales with available RAM
•Prevents massive KV cache allocation that caused slow startup
•8GB machines: 4096 context, 16GB: 8192, 32GB: 16384, 64GB+: 32768
Bug Fix
•Fixed: App window now shows immediately on startup; model loads in background
•Previously the window was blocked until the model finished loading
•Especially noticeable on 8GB RAM machines with fresh installs
Bug Fix
•Fixed: Update download progress now displays correctly when clicking "Download" for app updates
What's New
Features
•Internationalization (i18n): Added support for English and Dutch languages
•File-based logging: All logs now saved to disk with daily rotation (max 7 days)
Improvements
•Cleaned up duplicate translation keys
•"Open logs" button now opens folder with actual log files
Default LLM models from this version onwards
| RAM | Model | Size | Speed | Vision |
|---|---|---|---|---|
| 8–15 GB | qwen3.5-4b-ud-q5kxl | 3.25 GB | — | |
| 16–23 GB | qwen3.5-27b-q3km | 12.6 GB | 16 tok/s | — |
| 24–31 GB | qwen3.5-27b-q3km | 12.6 GB | 16 tok/s | ✓ |
| 32–63 GB | qwen3.5-35b-a3b-ud-q4km | 19.8 GB | 32 tok/s | ✓ |
| 64 GB+ | qwen3.5-122b-a10b-mxfp4 | 63 GB | 32 tok/s | ✓ |
Reference speed measured on Apple Mac Studio M1 Ultra. Actual speed depends on your hardware.
qwen3.5-4b-ud-q5kxl
3.25 GB
qwen3.5-27b-q3km
12.6 GB
qwen3.5-27b-q3km
12.6 GB
qwen3.5-35b-a3b-ud-q4km
19.8 GB
qwen3.5-122b-a10b-mxfp4
63 GB
ThinkTower v0.92.0
Downloads
•Apple Silicon (M1/M2/M3): `ThinkTower-0.92.0-arm64.dmg`
•Intel Mac: `ThinkTower-0.92.0-x64.dmg`
Default LLM models from this version onwards
| RAM | Model | Size | Speed |
|---|---|---|---|
| 8 GB | ministral-8b | 4.6 GB | 64 tok/s |
| 16 GB | qwen3.5-27b-q3km | 12.6 GB | 16 tok/s |
| 24 GB | qwen3.5-35b-a3b-ud-q4km | 19.8 GB | 32 tok/s |
| 48 GB | qwen3.5-35b-a3b-ud-q4km | 19.8 GB | 32 tok/s |
| 64 GB+ | qwen3.5-122b-a10b-mxfp4 | 63 GB | 32 tok/s |
Reference speed measured on Apple Mac Studio M1 Ultra. Actual speed depends on your hardware.
ministral-8b
4.6 GB
qwen3.5-27b-q3km
12.6 GB
qwen3.5-35b-a3b-ud-q4km
19.8 GB
qwen3.5-35b-a3b-ud-q4km
19.8 GB
qwen3.5-122b-a10b-mxfp4
63 GB
What's New
•Workmates now appear in the sidebar with accordion-style skill expansion
•Removed redundant workmate toolbar from welcome screen
•Various stability improvements
Downloads
•Apple Silicon (M1/M2/M3): ThinkTower-0.91.0-Mac-arm64.dmg
•Intel Mac: ThinkTower-0.91.0-Mac-x64.dmg
ThinkTower v0.90.5
Downloads
•Apple Silicon (M1/M2/M3): `ThinkTower-0.90.5-arm64.dmg`
•Intel Mac: `ThinkTower-0.90.5-x64.dmg`
ThinkTower v0.90.4
Downloads
•Apple Silicon (M1/M2/M3): `ThinkTower-0.90.4-arm64.dmg`
•Intel Mac: `ThinkTower-0.90.4-x64.dmg`
ThinkTower v0.90.3
Downloads
•Apple Silicon (M1/M2/M3): `ThinkTower-0.90.3-Mac-arm64.dmg`
•Intel Mac: `ThinkTower-0.90.3-Mac-x64.dmg`
•Windows: `ThinkTower-0.90.3-Windows-Setup.exe`
ThinkTower v0.90.2
Bugfixes
•Fixed license server connection issue
•Fixed native module compatibility on macOS
Downloads
macOS:
•Apple Silicon (M1/M2/M3/M4): ThinkTower-0.90.2-Mac-arm64.dmg
•Intel Mac: ThinkTower-0.90.2-Mac-x64.dmg
Windows:
•Windows 10/11: ThinkTower-0.90.2-Windows-Setup.exe
ThinkTower v0.90.1
Bugfix
•Fixed native module compatibility issue on macOS (better-sqlite3)
Downloads
macOS:
•Apple Silicon (M1/M2/M3/M4): ThinkTower-0.90.1-Mac-arm64.dmg
•Intel Mac: ThinkTower-0.90.1-Mac-x64.dmg
Windows:
•Windows 10/11: ThinkTower-0.90.1-Windows-Setup.exe
ThinkTower v0.90.0
Downloads
macOS:
•Apple Silicon (M1/M2/M3/M4): ThinkTower-0.90.0-Mac-arm64.dmg
•Intel Mac: ThinkTower-0.90.0-Mac-x64.dmg
Windows:
•Windows 10/11: ThinkTower-0.90.0-Windows-Setup.exe
Default LLM models from this version onwards
| RAM | Model | Size | Speed |
|---|---|---|---|
| tot 8 GB | ministral-8b | 4.6 GB | 64 tok/s |
| 24–47 GB | qwen3-30b | ~18 GB | 28 tok/s |
| 48+ GB | qwen3-80b | ~45 GB | 20 tok/s |
Reference speed measured on Apple Mac Studio M1 Ultra. Actual speed depends on your hardware.
ministral-8b
4.6 GB
qwen3-30b
~18 GB
qwen3-80b
~45 GB