When a new model enters AIDeskPro, we test it properly. Not on lab benchmarks - on our platform, on real conversations, on the agentic tasks our customers use every day.
GLM-5.2 arrived recently. The verdict is clear: it is our new default model for agentic tasks.
Scaleway brings it to Europe in 10 days
GLM-5.2 is an open-weight model developed by Z.ai. Scaleway, the French sovereign cloud provider, made it available just 10 days after release - among the first providers with European infrastructure to offer it (alongside Nebius Token Factory, which supported it from day zero).
All inference runs on Scaleway infrastructure in Europe. Data never leaves European servers, no third-party routing, full GDPR compliance. It is not a European model - but it runs and lives entirely in Europe.
It is built explicitly for the agentic era: configurable reasoning, no mid-task stalls, native handling of complex pipelines and multi-step workflows.
On FrontierSWE - the most rigorous benchmark for long-horizon coding tasks - GLM-5.2 sits within 1% of Claude Opus 4.8, outperforming GPT-5.5 and Gemini 3.1 Pro. It is the best-performing open-weight model across all major coding benchmarks. MIT licence, no geographic restrictions.
Why we changed our default model
On AIDeskPro we used Claude Sonnet 4.6 as our reference for agentic tasks. Good quality-to-price ratio, reliable. But when we tested GLM-5.2 in production, the comparison was straightforward:
| GLM-5.2 | Claude Sonnet 4.6 | |
|---|---|---|
| Input | €1.80 / M tokens | $3.30 / M tokens |
| Output | €5.50 / M tokens | $16.50 / M tokens |
| Infrastructure | Europe (Scaleway) | Europe (dedicated endpoint) |
| Licence | MIT open-weight | Proprietary |
Around 45% savings on cost per token (accounting for the dollar/euro exchange rate). And from our tests, GLM-5.2 is also faster and performs better on complex agentic tasks. This is not a trade-off: it is an upgrade across the board.
PPP: Performance, Price, Privacy
In our internal tests we summarised the result in three letters.
Performance. Faster responses, more accurate on multi-step tasks, less prone to mid-reasoning stalls on long tasks.
Price. Around 45% less than Claude Sonnet with European endpoints, accounting for the dollar/euro exchange rate. At enterprise volumes, the difference is very significant.
Privacy. Sovereign European infrastructure on Scaleway. Data does not leave European servers, no third-party routing, full GDPR compliance.
Available now in AIDeskPro
GLM-5.2 is already available in AIDeskPro and is now our default recommendation for agentic workflows. Models evolve quickly and we keep testing - but today, for this type of task, GLM-5.2 is the best in our catalogue for the combination of quality, speed and price.