Claude Sonnet 4.6 Tested - It's Incredibly Slow and Makes Dumb Mistakes (But Also Amazing?)

Video Description

Anthropic just dropped Claude Sonnet 4.6 and honestly? It's the most frustrating model I've ever tested. On one hand, Claude Sonnet 4.6 produces some of the best outputs I've seen from any AI. The architect website is stunning. The Duolingo clone actually works (unlike GLM5 and Minimax M2.5). The Vampire Survivor game has bosses and increasing difficulty. On the other hand, Claude Sonnet 4.6 makes mistakes that even cheaper models don't make. "Free neighborhoods" on a Rome travel page? Layout bugs everywhere? A save button on items that are already saved? And it's PAINFULLY slow – 10 to 15+ minutes per generation, even on medium thinking. This Claude Sonnet 4.6 review covers my full coding test: 6 real tasks, same prompts as my GLM5 and Kimi K2.5 tests. Some results are the best I've ever seen. Others are embarrassingly bad. Watch to see exactly where Claude Sonnet 4.6 shines and where it completely fails. ⏱️ TIMESTAMPS 0:00 - Claude Sonnet 4.6 First Impressions 0:26 - Test 1: Architect Website Homepage 2:04 - Test 2: SimCity Clone 4:44 - Test 3: Vampire Survivor Clone 6:22 - Test 4: Tinder Clone for Clothes 8:27 - Test 5: Rome Budget Travel Page (Agentic) 13:55 - The Problems: Stupid and Slow 16:32 - Test 6: Duolingo Clone 18:56 - Final Verdict: Frustrating but Capable 📊 CLAUDE SONNET 4.6 CODING TEST RESULTS | Test | Result | Notes | |------|--------|-------| | Architect Website | ⚠️ Mixed | Stunning design, broken hero | | SimCity Clone | ⚠️ Mixed | Great simulation, looks empty | | Vampire Survivor | ✅ Pass | Bosses, difficulty scaling, fun | | Tinder Clone | ⚠️ Mixed | Good design, stupid UX errors | | Rome Budget Page | ⚠️ Mixed | Amazing sections + terrible sections | | Duolingo Clone | ✅ Pass | Best version I've seen! | **Claude Sonnet 4.6 Score: 2 Clear Passes, 4 Mixed Results** ⏱️ THE SPEED PROBLEM Claude Sonnet 4.6 generation times (medium thinking): - Fashion App: 10+ minutes - Landing Page: 15-20 minutes (first version) + 8 minutes (images) - Duolingo Clone: 15 minutes 36 seconds This is on MEDIUM thinking. The model spends enormous amounts of tokens thinking, but still makes obvious mistakes. Why? 🤔 THE FRUSTRATING CONTRADICTION **Where Claude Sonnet 4.6 is AMAZING:** - Website hero sections and visual design - Image sourcing (best I've seen from any Claude model) - Complex game logic (Vampire Survivor with bosses) - Duolingo clone functionality and sounds - SimCity simulation mechanics **Where Claude Sonnet 4.6 is STUPID:** - "Free neighborhoods" – nonsensical content - 4th column below 3 columns (layout bug everywhere) - Save button on already-saved items - No images on "best landing page" until asked - Inconsistent quality within the same page The same model that creates stunning designs also writes "free neighborhoods" on a travel page. How? 🆚 CLAUDE SONNET 4.6 VS OTHER MODELS **Duolingo Clone Comparison:** - Claude Sonnet 4.6: ✅ Best version (sounds, levels, works perfectly) - Kimi K2.5: ✅ Works well - GLM5: ❌ Failed - Minimax M2.5: ❌ Failed **Speed Comparison:** - GLM5: Fast - Kimi K2.5: Slow - Claude Sonnet 4.6: Painfully slow (10-20 minutes) **Consistency:** - Kimi K2.5: Most consistent - GLM5: Good consistency - Claude Sonnet 4.6: Wild inconsistency 💡 KEY FINDINGS FROM CLAUDE SONNET 4.6 REVIEW 1. **Thinking time ≠ Quality** – Despite spending 15+ minutes thinking, Claude Sonnet 4.6 still makes obvious mistakes 2. **Section-by-section quality variance** – The same page can have stunning sections and amateur-looking sections 3. **Can't trust autonomous decisions** – When Claude Sonnet 4.6 has to decide what content to include, it sometimes produces nonsense 4. **Best Duolingo clone** – Finally a model that passes this test with flying colors 5. **Possibly the "Pony Alpha" model** – The outputs look very similar to the stealth model that was pulled 🔧 CLAUDE SONNET 4.6 TEST SETUP - Platform: Claude Code - Thinking Level: Medium - Same prompts as GLM5, Kimi K2.5, Minimax M2.5 tests - No skills or plugins used ⏳ WHAT'S NEXT Head-to-head comparison coming: - Claude Sonnet 4.6 vs Claude Opus 4.6 (is Opus faster?) - Claude Sonnet 4.6 vs GLM5 vs Kimi K2.5 - Complex backend tasks - Speed benchmarks Subscribe to see which model actually wins. 🏷️ TAGS #ClaudeSonnet46 #ClaudeSonnet #Anthropic #Claude46 #SonnetCoding #ClaudeCode #AICoding #ClaudeReview #Sonnet46Review #VibeCoding #AIComparison #CodingTest #ClaudeAI #AnthropicAI #Sonnet46Test #ClaudeCodingTest #AIReview #Claude46Coding — Subscribe to AIchievable for honest AI reviews. When a model frustrates me, I tell you. When it's amazing, I tell you that too. Claude Sonnet 4.6 is somehow both.