Gemma 3 27B IT

Roboflow's Playground page provides independent specification corroboration for Gemma 3 27B: roughly 27 billion parameters, multimodal (text and image input, text output), a 128,000-token context window, 8,192-token max output, instruction-tuned for chat/reasoning/summarization, structured outputs and function calling Roboflow's vision-focused benchmark gives a 58.21% pass rate across 67 visual understanding tasks (ranked 51 of 77), an average response time of 33.60 seconds per task, and roughly $0.080 in inference cost across 65 inferences in the past 30 days. The page explicitly notes the Vision Evals section is legacy and the cur

The OpenRouter marketplace listing for Google's Gemma 3 27B IT shows NovitaAI as one of five hosting providers, with provider-specific data: $0.119 per million input tokens, $0.20 per million output tokens, 0.55s latency, 32 tokens/sec throughput, 94.35% uptime, and roughly 9.9% token share. Effective 30-day weighted-a The listing confirms Gemma 3 27B's core capabilities: multimodal vision-language input with text output, up to a 262K context window as reported by OpenRouter (versus Google's documented 128K native, a noted discrepancy), 140+ language support, structured outputs, and function calling, with a March 12, 2025 release and