Releasesby GlyphEdit the system prompt
You can now edit the system prompt when you chat with a model in the Playground . Each model remembers your system prompt. To go back to the default prompt, click Reset .
Also shipped Oct 13, 2026Modular in October 2026 →
What's included:
- Three training methods: Supervised fine-tuning (SFT) , direct preference optimization (DPO) , and reinforcement learning with group relative policy optimization (GRPO) . Each method trains a low-rank adapter (LoRA) or the full model weights. Custom loss functions : Train with any objective you compute in PyTorch from the model's log-probabilities.
The AI.AGG function is generally available (GA)
The AI.AGG function is generally available (GA). You can use this function to semantically aggregate unstructured input data based on natural language instructions.
BigQuery generative AI functions now support the gemini-4-argon model
BigQuery generative AI functions now support the gemini-4-argon model. This model is generally available .
Gemini 4 Argon in CodeMender (Preview)
CodeMender now supports Gemini 4 Argon. To use it, pass --model gemini-4-argon to cm find , cm verify , or cm fix . Gemini 3.8 Flash remains the default model. For more information, see Specifying the model .
Gemini 4 Argon and Gemini 4 Argon Cyber Permissive
Gemini 4 Argon ( gemini-4-argon ) is available in General Availability (GA) . Gemini 4 Argon is Google's frontier model that excels at deep work in complex, real-world tasks across domains including coding, finance, legal, cybersecurity, and science.
On-demand state repartitioning is now generally available
On-demand state repartitioning is now generally available. You can resize the number of shuffle partitions for a stateful Structured Streaming query or Lakeflow pipeline without losing checkpoint state. See On-demand state repartitioning for stateful streaming queries .
More from ModularEvery Modular release →
Set a monthly spend limit
You can now cap how much your organization spends each billing cycle with a monthly spend limit . When your organization reaches its limit, API requests return a 401 error until the next billing cycle starts or an admin raises or removes the limit.
Use the Playground without an API key
You don't need to create an API key to try a model anymore. Open the Playground , select a shared model, and start running inference. Playground usage still bills to your organization.
Choose which emails you get
Your profile page now has email preferences for billing alerts, tips, and marketing.
New model: MiniMax M3.1 Flash
MiniMax M3.1 Flash is now available on shared endpoints with options for dedicated deployments.
Start where you left off
Your work in the Playground now carries over between visits. Previous chats for each model are saved automatically and listed in the right sidebar. Click any conversation to pick back up on it. Each model now remembers the inference settings you last used, such as max tokens, temperature, and reasoning mode.
See more accurate usage numbers
A model endpoint shows more accurate usage: We now report cached input tokens in model usage. FLUX megapixels now appear in model usage.
Auto-reload your credits
Now you can set up automatic credit reloading in the Billing page. Enable this so you won't run out of credits in the middle of a job. This applies to users on the Starter plan.
GLM 5.3 dedicated deployment support
You can now deploy GLM 5.3 as a dedicated deployment in Modular Cloud. For the full list of shared endpoints and dedicated deployment options, see Supported models .