Google DeepMind released Gemini 3.6 Flash and Gemini 3.5 Flash-Lite on July 21, 2026. Both models feature enhanced token efficiency and reduce task completion time by nearly 50% compared to their respective predecessors. Gemini 3.6 Flash matches the intelligence level of Gemini 3.5 Flash, scoring 50 on the Artificial Analysis Intelligence Index, while Gemini 3.5 Flash-Lite achieves a significant 11-point gain, reaching a score of 36.

Gemini 3.6 Flash records an average time per task of 1.3 minutes, a 50% reduction from the 2.7 minutes required by Gemini 3.5 Flash. The model's cost per task decreased by approximately 18%, dropping from $0.59 to $0.50, supported by new pricing of $1.50 per 1M input tokens and $7.50 per 1M output tokens. It retains a 1M context window (total amount of information a model considers at once) and provides multimodal input capabilities including text, image, video, and speech.

Gemini 3.5 Flash-Lite achieves a 0.6-minute average time per task, down from 1.0 minute in the 3.1 iteration. Despite higher processing speed, the average cost per task rose from $0.04 to $0.09 due to updated pricing of $0.30 per 1M input tokens and $2.50 per 1M output tokens. Performance gains for the Flash-Lite variant are notably driven by improvements in agentic evaluations (tasks where an AI acts independently to complete goals), specifically in benchmarks such as TerminalBench v2.1, which saw a 22.5-point increase. Both new models continue to offer a 90% discount for cached input tokens.

Google DeepMind released Gemini 3.6 Flash and Gemini 3.5 Flash-Lite on July 21, 2026. Both models feature enhanced token efficiency and reduce task completion time by nearly 50% compared to their respective predecessors. Gemini 3.6 Flash matches the intelligence level of Gemini 3.5 Flash, scoring 50 on the Artificial Analysis Intelligence Index, while Gemini 3.5 Flash-Lite achieves a significant 11-point gain, reaching a score of 36.

Gemini 3.6 Flash records an average time per task of 1.3 minutes, a 50% reduction from the 2.7 minutes required by Gemini 3.5 Flash. The model's cost per task decreased by approximately 18%, dropping from $0.59 to $0.50, supported by new pricing of $1.50 per 1M input tokens and $7.50 per 1M output tokens. It retains a 1M context window (total amount of information a model considers at once) and provides multimodal input capabilities including text, image, video, and speech.

Gemini 3.5 Flash-Lite achieves a 0.6-minute average time per task, down from 1.0 minute in the 3.1 iteration. Despite higher processing speed, the average cost per task rose from $0.04 to $0.09 due to updated pricing of $0.30 per 1M input tokens and $2.50 per 1M output tokens. Performance gains for the Flash-Lite variant are notably driven by improvements in agentic evaluations (tasks where an AI acts independently to complete goals), specifically in benchmarks such as TerminalBench v2.1, which saw a 22.5-point increase. Both new models continue to offer a 90% discount for cached input tokens.