At a glance
WHAT IT’S REALLY ABOUT
Testing GLM 5.2 as low-cost Opus alternative in coding
- GLM 5.2 is presented as an open-weight, text-only model with modern tooling features (reasoning mode, function calling, caching, structured output) and a 1M-token context window.
- Benchmark positioning suggests GLM 5.2 competes near Claude Opus and GPT-class models on coding suites, motivating hands-on validation rather than relying on leaderboard claims.
- Claire shows practical setup paths using OpenRouter as a hosted provider, including a Cursor-specific base URL nuance and Claude Code environment/config changes.
- Live tests show strong codebase understanding and surprisingly solid HTML/communication output, plus generally acceptable design iteration when anchored to an existing design system.
- The longest autonomous run successfully pulled Sentry/Vercel signals and produced a prioritized fix plan, though the model temporarily struggled with TypeScript/React generation speed and reliability; total spend was about $3.36 for ~6M tokens with high cache hit rate.
IDEAS WORTH REMEMBERING
5 ideasOpen-weight can mean control and flexibility, not necessarily “free.”
Because model weights are downloadable, you can self-host, fine-tune, and avoid vendor lock-in, but licensing and inference costs still matter depending on provider and usage.
GLM 5.2’s feature parity makes it viable in modern coding stacks.
Despite being text-to-text only, it supports the workflow primitives developers rely on (reasoning mode, tool/function calls, streaming, caching, structured outputs, MCP), plus a large context window for repo-scale tasks.
Setup friction is real—small URL details can block adoption.
In Cursor, Claire had to place the OpenRouter key into the “OpenAI API key” field and use the specific base URL `openrouter.ai/api/v1/cursor`; missing `/cursor` was the undocumented gotcha.
Repo orientation and architecture narration were strong early signals.
When asked to explore the ChatPRD codebase and summarize architecture/recent work, the model responded quickly with an accurate, useful overview—suggesting good context use and software-engineering intuition.
GLM 5.2 is notably good at HTML “explainers,” not just code fixes.
Turning an architecture/roadmap summary into a presentable HTML page produced a credible artifact with sensible structure, product pillars, and roadmap items—useful for bridging agent output to human review.
WORDS WORTH SAVING
5 quotesWhat if I told you you could get Opus-level reasoning at a fraction of the cost?
— Claire Vo
This is our first of many reviews of open-weight and open source models to see if we should all be paying the tax to Anthropic and OpenAI, or if we can run these models locally and get the same results.
— Claire Vo
I could not find anything for a really long time that told me it had to be /cursor, but it is /cursor, and you need to toggle that change on.
— Claire Vo
Oh my God, guys, it really is having trouble writing JavaScript right now.
— Claire Vo
I spent $3.36 on about 6 million tokens.
— Claire Vo
High quality AI-generated summary created from speaker-labeled transcript.
