[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-glm-5-3-coding-gains-post-training-en":3,"article-related-glm-5-3-coding-gains-post-training-en":31,"series-industry-3ba7bf6c-88c7-4563-b5af-243565c5523c":78},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":24,"views":28,"created_at":29,"published_at":30,"topic_cluster_id":11},"3ba7bf6c-88c7-4563-b5af-243565c5523c","glm-5-3-coding-gains-post-training-en","GLM-5.3’s coding gains came from post-training","\u003Cp>Where did GLM-5.3’s \u003Ca href=\"\u002Fnews\u002Fgemini-3-7-flash-launch-coding-gains-en\">coding gains\u003C\u002Fa> come from?\u003C\u002Fp>\u003Cp data-speakable=\"summary\">GLM-5.3 improved coding mostly through post-training, not a new base model.\u003C\u002Fp>\u003Ch2>1. Post-training, not a new foundation model\u003C\u002Fh2>\u003Cp>The core claim in the GLM-5.3 story is simple: the model’s coding jump did not come from retraining the foundation model from scratch. Instead, the gains came after pretraining, through targeted post-training that shaped how the model reasons, follows instructions, and writes code.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786773762056-613t.png\" alt=\"GLM-5.3’s coding gains came from post-training\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>That matters because it changes the engineering question. If you are comparing model upgrades, you cannot assume a bigger base model is the only path to better coding. Sometimes the better move is to keep the base intact and spend effort on the stages after it.\u003C\u002Fp>\u003Cul>\u003Cli>Base model: unchanged\u003C\u002Fli>\u003Cli>Improvement source: post-training\u003C\u002Fli>\u003Cli>Outcome: stronger coding behavior\u003C\u002Fli>\u003Cli>Practical lesson: inspect training pipeline, not just parameter count\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>2. Coding-focused supervision\u003C\u002Fh2>\u003Cp>One likely source of the improvement is code-heavy supervision, where the model is trained on tasks that look more like real developer work. That can mean bug fixes, function completion, test generation, refactoring, and multi-step problem solving rather than generic text prediction.\u003C\u002Fp>\u003Cp>This kind of tuning helps because coding is less about fluent prose and more about correctness under constraints. A model can look smart in chat and still fail at imports, edge cases, or \u003Ca href=\"\u002Ftag\u002Fapi\">API\u003C\u002Fa> usage. Post-training on code tasks narrows that gap.\u003C\u002Fp>\u003Cul>\u003Cli>Bug fixing and repair\u003C\u002Fli>\u003Cli>Unit test generation\u003C\u002Fli>\u003Cli>API usage and call patterns\u003C\u002Fli>\u003Cli>Multi-file code changes\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>3. Better reasoning traces\u003C\u002Fh2>\u003Cp>Another likely ingredient is reasoning-oriented training. If a model learns to break a coding problem into steps before producing an answer, it can do better on tasks that require planning, dependency tracking, or debugging across several files.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786773762722-ls1i.png\" alt=\"GLM-5.3’s coding gains came from post-training\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>In practice, this means the model is not only learning syntax. It is also learning how to inspect a problem, form a sequence, and avoid jumping straight to a brittle answer. For coding assistants, that can be more valuable than raw \u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> gains.\u003C\u002Fp>\u003Ccode>Example workflow:\n1. Read error message\n2. Identify likely subsystem\n3. Check related code paths\n4. Propose patch\n5. Suggest validation test\u003C\u002Fcode>\u003Ch2>4. Agent-style evaluation and feedback\u003C\u002Fh2>\u003Cp>The article points to a broader trend in model development: train and evaluate models the way they will actually be used. For coding, that means letting the model operate in workflows where it can plan, inspect outputs, revise its answer, and respond to feedback.\u003C\u002Fp>\u003Cp>This is especially relevant for code agents. A model that can draft code once is useful, but a model that can improve its own draft after seeing compiler errors or test failures is much more valuable. Post-training can reward that loop.\u003C\u002Fp>\u003Cul>\u003Cli>Generate code\u003C\u002Fli>\u003Cli>Run checks or tests\u003C\u002Fli>\u003Cli>Read failures\u003C\u002Fli>\u003Cli>Revise the patch\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>5. Why this matters for model buyers\u003C\u002Fh2>\u003Cp>GLM-5.3 is a reminder that benchmark charts can hide where performance really came from. Two models with similar base architectures can behave very differently if one got stronger post-training for coding, instruction following, or \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> behavior.\u003C\u002Fp>\u003Cp>That means buyers should ask about the training recipe, not just the model name. The best coding model for a team may be the one with the better post-training stack, even if its base model is not the newest or largest.\u003C\u002Fp>\u003Cul>\u003Cli>Ask whether the base model changed\u003C\u002Fli>\u003Cli>Ask what post-training data was used\u003C\u002Fli>\u003Cli>Ask how coding was evaluated\u003C\u002Fli>\u003Cli>Ask whether the model was tuned for agent workflows\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>How to decide\u003C\u002Fh2>\u003Cp>If you care about raw coding help, prioritize models with explicit code supervision and agent-style evaluation. If you care about general chat quality, a stronger base model may still matter more.\u003C\u002Fp>\u003Cp>For teams buying or benchmarking models, GLM-5.3 is a useful warning: the best coding results may come from post-training choices that are easy to miss if you only look at parameter counts or headline scores.\u003C\u002Fp>","4 post-training moves explain how GLM-5.3 improved coding without changing its base model.","thenewstack.io","https:\u002F\u002Fthenewstack.io\u002Fglm-5-3-post-training-coding\u002F",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786773762056-613t.png","industry","en","54ada8df-3058-4dd1-94ef-e7e78f611e22",[17,18,19,20,21,22,23],"GLM-5.3","post-training","coding model","AI agents","code supervision","reasoning","model evaluation",[25,26,27],"GLM-5.3’s coding gains came from post-training, not a new base model.","Code supervision and reasoning-oriented tuning can matter more than model size.","Buyers should inspect the training recipe, not just benchmark numbers.",0,"2026-08-15T06:02:18.768877+00:00","2026-08-15T06:02:18.761+00:00",{"tags":32,"relatedLang":37,"relatedPosts":41},[33,35],{"name":19,"slug":34},"coding-model",{"name":20,"slug":36},"ai-agents",{"id":15,"slug":38,"title":39,"language":40},"glm-5-3-coding-gains-post-training-zh","GLM-5.3 編碼提升，重點在後訓練","zh",[42,48,54,60,66,72],{"id":43,"slug":44,"title":45,"cover_image":46,"image_url":46,"created_at":47,"category":13},"b268ffd0-3db8-498e-a90a-56f06614ad4e","anthropic-data-center-push-big-capital-backing-en","Anthropic’s data center push now has big capital backing","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786752163701-frj9.png","2026-08-15T00:02:21.497479+00:00",{"id":49,"slug":50,"title":51,"cover_image":52,"image_url":52,"created_at":53,"category":13},"71d8e28f-7271-4097-aa4c-e0f6abb387c0","dailyarxiv-arxiv-keyword-paper-feed-en","DailyArxiv turns arXiv keywords into a daily paper feed","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786716167937-eko7.png","2026-08-14T14:02:23.231042+00:00",{"id":55,"slug":56,"title":57,"cover_image":58,"image_url":58,"created_at":59,"category":13},"dffb9b37-a7d8-4c03-92ab-2679e2b25179","claude-vs-chatgpt-2026-is-claude-better-en","Claude vs ChatGPT (2026): Is Claude Better?","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786710770818-v38r.png","2026-08-14T12:32:29.804007+00:00",{"id":61,"slug":62,"title":63,"cover_image":64,"image_url":64,"created_at":65,"category":13},"5b01c39e-b627-458f-a6fc-b8a349ed0cf3","cbdc-governance-standards-stakeholder-roles-en","CBDC Governance: Standards and Stakeholder Roles","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786672975339-8lfe.png","2026-08-14T02:02:29.634243+00:00",{"id":67,"slug":68,"title":69,"cover_image":70,"image_url":70,"created_at":71,"category":13},"ed5cf5b1-9fe8-4ac8-b1ce-99017760ac54","invisible-ai-watermarks-right-move-claude-en","Invisible AI watermarks are the right move for Claude","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786665773072-vj2y.png","2026-08-14T00:02:28.15401+00:00",{"id":73,"slug":74,"title":75,"cover_image":76,"image_url":76,"created_at":77,"category":13},"fae47509-4ad5-4546-8936-e7e1fe24c0df","pgvector-enough-small-datasets-not-default-en","pgvector is enough for small datasets, not your default","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786656761311-ngh2.png","2026-08-13T21:32:18.817428+00:00",[79,84,89,94,99,104,109,114,119,124],{"id":80,"slug":81,"title":82,"created_at":83},"d35a1bd9-e709-412e-a2df-392df1dc572a","ai-impact-2026-developments-market-en","AI's Impact in 2026: Key Developments and Market Shifts","2026-03-25T16:20:33.205823+00:00",{"id":85,"slug":86,"title":87,"created_at":88},"5ed27921-5fd6-492e-8c59-78393bf37710","trumps-ai-legislative-framework-en","Trump's AI Legislative Framework: What's Inside?","2026-03-25T16:22:20.005325+00:00",{"id":90,"slug":91,"title":92,"created_at":93},"e454a642-f03c-4794-b185-5f651aebbaca","nvidia-gtc-2026-key-highlights-innovations-en","NVIDIA GTC 2026: Key Highlights and Innovations","2026-03-25T16:22:47.882615+00:00",{"id":95,"slug":96,"title":97,"created_at":98},"0ebb5b16-774a-4922-945d-5f2ce1df5a6d","claude-usage-diversifies-learning-curves-en","Claude Usage Diversifies, Learning Curves Emerge","2026-03-25T16:25:50.770376+00:00",{"id":100,"slug":101,"title":102,"created_at":103},"69934e86-2fc5-4280-8223-7b917a48ace8","openclaw-ai-commoditization-concerns-en","OpenClaw's Rise Raises Concerns of AI Model Commoditization","2026-03-25T16:26:30.582047+00:00",{"id":105,"slug":106,"title":107,"created_at":108},"b4b2575b-2ac8-46b2-b90e-ab1d7c060797","google-gemini-ai-rollout-2026-en","Google's Gemini AI Rollout Extended to 2026","2026-03-25T16:28:14.808842+00:00",{"id":110,"slug":111,"title":112,"created_at":113},"6e18bc65-42ae-4ad0-b564-67d7f66b979e","meta-llama4-fabricated-results-scandal-en","Meta's Llama 4 Scandal: Fabricated AI Test Results Unveiled","2026-03-25T16:29:15.482836+00:00",{"id":115,"slug":116,"title":117,"created_at":118},"bf888e9d-08be-4f47-996c-7b24b5ab3500","accenture-mistral-ai-deployment-en","Accenture and Mistral AI Team Up for AI Deployment","2026-03-25T16:31:01.894655+00:00",{"id":120,"slug":121,"title":122,"created_at":123},"5382b536-fad2-49c6-ac85-9eb2bae49f35","mistral-ai-high-stakes-2026-en","Mistral AI: Facing High Stakes in 2026","2026-03-25T16:31:39.941974+00:00",{"id":125,"slug":126,"title":127,"created_at":128},"9da3d2d6-b669-4971-ba1d-17fdb3548ed5","cursors-meteoric-rise-pressures-en","Cursor's Meteoric Rise Faces Industry Pressures","2026-03-25T16:32:21.899217+00:00"]