Trying out Gemini 3 Pro with audio transcription and a new pelican benchmark
4 days ago
- #Benchmark
- #AI
- #Gemini 3 Pro
- Google released Gemini 3 Pro, an upgrade to Gemini 2.5, matching leading rival models.
- Gemini 3 Pro maintains the same knowledge cutoff (January 2025) and supports 1M input tokens, 64K output tokens, with multimodal inputs (text, images, audio, video).
- Benchmarks show Gemini 3 Pro slightly outperforming Claude 4.5 Sonnet and GPT-5.1 in standard tests.
- Pricing for Gemini 3 Pro is higher than Gemini 2.5 but cheaper than Claude Sonnet 4.5.
- Gemini 3 Pro was tested with a complex image, generating accurate alt text for screen readers.
- Audio transcription test with a city council meeting was partially successful, with timestamp inaccuracies.
- Gemini 3 Pro introduced a 'thinking level' feature (low/high), demonstrated via a pelican SVG benchmark.
- A new, more detailed pelican benchmark was introduced, testing AI's ability to generate detailed SVGs.