[Google Releases Gemini 3.6 Flash, Gemini 4 Pre-Training Also Underway] According to monitoring by Beating Insights, Google has released Gemini 3.6 Flash, focusing on programming, multimodal capabilities, and multi-step agent workflows. The new model reduces reasoning steps, tool calls, and execution loops. In the Artificial Analysis Index, its token output usage is 17% lower than Gemini 3.5 Flash. The API input price remains $1.5 per million tokens, while the output price has dropped from $9 to $7.5. Performance has also improved: DeepSWE increased from 37% to 49%, MLE Bench from 49.7% to 63.9%, and OSWorld-Verified from 78.4% to 83%. The model is now officially available, supporting a 1 million token context and up to 64,000 token output. Its biggest change is enabling agents to take fewer detours, use fewer tokens, and reduce costs for long tasks. Google also announced that Gemini 3.5 Pro is being opened for testing with partners and will be fully launched once preparations are complete. Pre-training for Gemini 4 has also begun, marking Google's most ambitious round of pre-training to date, though no release date or further details have been disclosed by the company. [Original Link]