Kimi K3 and GLM 5.3 Flash are now generally available
Developer Relations
In short:
- Kimi K3 and GLM 5.3 Flash are now generally available (stable) on Berget AI.
- GLM 5.2 is deprecated — with end-of-life (EOL) on September 18.
- GLM 4.7 (EOL September 4), along with GPT-OSS 120B and Llama 3.3 70B (EOL September 11) also retire this month.
Kimi K3 and GLM 5.3 Flash become generally available
Many of you have already been relying on both Kimi K3 and GLM 5.3 Flash for a while since we released them for public evaluation shortly after their weights were made public.
During that time we've been able to test and benchmark them under real workloads, and we're now confident that they're fit for production use.
Both models now run on sovereign Swedish infrastructure. Together they cover very different needs. Kimi K3 handles the heaviest reasoning, while GLM 5.3 Flash handles a wide range of agentic workloads at a lower price.
The catalog at a glance
| Kimi K3 | GLM 5.3 Flash | |
|---|---|---|
| Model ID | moonshotai/Kimi-K3 | zai-org/GLM-5.3-Flash |
| Modalities | Text + vision | Text + vision |
| Context window | 320,000 | 320,000 |
| Price (in/out per M tokens) | €3.00 / €15.00 | €0.25 / €0.50 |
| Built for | Maximum-capability workloads | High-volume, cost-sensitive workloads |
GLM 5.2 reaches end-of-life
When we launched GLM 5.2 in June, it was the leading open-weights model for general intelligence, and a lot of production traffic has run on it since. With Kimi K3 and GLM 5.3 Flash reaching general availability, we'll be sunsetting GLM 5.2 from our model catalogue. The last day of service is 17 September 2026.
For most GLM 5.2 users the move is to 5.3 Flash: same family and profile for a fraction of the price. If you used 5.2 for the hardest reasoning tasks, move that traffic to Kimi K3.
Why we're deprecating models
GLM 5.2 isn't the only retirement this month. GLM 4.7 retires on September 4, followed by GPT-OSS 120B and Llama 3.3 70B on September 11.
Some of these models have been with us from the very start. With the rapid improvements in open-weights model during this year, we're freeing up resources to be able to evaluate new models as improving the capacity and reliability of the ones we already have.
If you have questions of need guidance on models to migrate to, contact us and we'll help you plan your migration.