GLM 5 Coding Test - One Was the Best I've Seen, One Completely Broke - GLM 5 Benchmark

Video Description

GLM 5 coding test GLM 5 just dropped and I ran my 6 standard coding tests – same prompts, same setup as my Kimi K2.5 video. The results? All over the place. One generation was the best I've ever seen from any AI model. The Tinder clone finally used full-size images – something no other model has done before. The SimCity clone tracks population AND jobs. But the Duolingo clone? Completely broken. Couldn't even submit answers. GLM 5 is fast, capable, and produces some incredible one-shot generations. But it's inconsistent. Some tests beat Kimi K2.5, others fell short. Watch to see exactly where GLM 5 shines and where it struggles. ⏱️ TIMESTAMPS 0:00 - GLM 5 Coding Test Introduction 0:30 - Test 1: Architect Website 1:40 - Test 2: Vampire Survivor Clone 3:31 - Test 3: Tinder Clone for Clothes 4:50- Test 4: SimCity Clone 6:04 - Test 5: Rome Budget Travel Page (Agentic) 9:23 - Test 6: Duolingo Clone 11:35 - Final Verdict: GLM 5 vs Kimi K2.5 📊 TEST RESULTS SUMMARY | Test | Result | Notes | |------|--------|-------| | Architect Website | ⚠️ Okay | Good style, lacking content | | Vampire Survivor | ✅ Pass | Great gameplay, auto-aim option | | Tinder Clone | ✅ Best Ever | First AI to use full-size images! | | SimCity Clone | ✅ Best Ever | Population + jobs tracking | | Rome Budget Page | ⚠️ Mixed | Chat failed, Agent mode great | | Duolingo Clone | ❌ Fail | Couldn't submit typed answers | **Final Score: 4/6 Passed, 1 Mixed, 1 Failed** 🏆 STANDOUT MOMENTS **Tinder Clone – Historic First:** GLM 5 is the FIRST AI model I've tested that displays images at full size in a feed app. Every other model crops or shrinks images. This is exactly what you need for a fashion/product app. Finally. **SimCity Clone – Best Generation:** - Population counter that actually works - Job tracking system - Buildings grow over time - Lights appear as city develops - Speed controls - One-shot generation **Duolingo Clone – Complete Fail:** - Typing answers doesn't work - Submit button unresponsive - Shows 100% even when broken - Second attempt still failed - Kimi K2.5 did this much better 💡 KEY FINDINGS **GLM 5 Strengths:** - Exceptional game development - Full-size image handling (finally!) - Complex simulations work well - Agent mode produces better results than chat - Very fast output speed **GLM 5 Weaknesses:** - Website content generation lacking - Form/input handling buggy - Chat mode less reliable than agent mode - Inconsistent across different task types **GLM 5 vs Kimi K2.5:** - Games: Similar quality, GLM faster - Websites: Kimi better design + content - Apps with inputs: Kimi more reliable - Agentic tasks: Both good, different styles - Speed: GLM significantly faster 🔍 PONY ALPHA COMPARISON I also compared GLM 5 outputs to the "stealth model" Pony Alpha: - Same app names generated - Completely different visual styles - Different code approaches - You decide if it's the same model... 🔧 TECHNICAL NOTES - Tests run in GLM 5 web interface - Agent mode used for agentic tasks - All games were one-shot generations - No external tools or plugins - Rumored 745B parameters (unconfirmed) ⏳ WHAT'S NEXT Waiting for GLM 5 coding plan to test in Claude Code. Then we can compare: - GLM 5 vs Opus 4.6 - Real backend implementations - Complex multi-file projects Subscribe to see those tests when they drop. 🏷️ TAGS #GLM5 #GLM5Coding #CodingTest #AIGames #VampireSurvivor #SimCity #Duolingo #AICoding #ZhipuAI #ChatGLM #KimiK25 #VibeCoding #AIComparison #GameDev #AIReview #CodingBattle #GLMvsClaude #GLMvsKimi — Subscribe to AIchievable for honest AI coding tests. When a new model drops, I test it the same day. No hype, just real results.