Anthropic, an AI startup, has emerged as a notable player in the field of AI models since the release of OpenAI’s ChatGPT. In March, Anthropic introduced its own AI model called Claude, which has proven to be a strong competitor to OpenAI’s GPT-3.5 and GPT-4. Alongside Claude, Anthropic also launched Claude Instant, a lighter, faster, and more affordable version of the model. Now, Anthropic has unveiled an upgraded version called Claude Instant 1.2, which leverages the advanced capabilities of Claude 2.0, the latest version of Claude released in July.
According to Anthropic, Claude Instant 1.2 has shown significant improvements in math, coding, reasoning, and safety compared to its predecessor. The model generates longer and more structured responses, thanks to the enhanced abilities of Claude 2.0. To validate its performance, Anthropic conducted benchmark evaluations such as the Codex evaluation and Grade-school math problems benchmark (GSM8k), which assess math and coding capabilities.
In both the Codex evaluation and GSM8k, Claude Instant 1.2 outperformed its previous version. It achieved a score of 58.7% in the Codex evaluation, surpassing the original’s score of 52.8%. Similarly, in GSM8k, Claude Instant 1.2 scored 86.7% compared to the original’s 80.9%. Although the newer model performed slightly better or worse than the older model in other benchmark exams, the differences were minimal.
Notably, the quality of answer output has also improved with Claude Instant 1.2. It exhibits decreased hallucinations and increased resistance to jailbreaking attempts, making it a safer model to use. A red-teaming evaluation confirmed the enhanced safety of Claude 1.2.
Businesses interested in accessing the new model can fill out an interest form to gain access. Developers, on the other hand, have the option to utilize the API, which is significantly more affordable than Claude 2.0.
Anthropic’s continued advancements in AI models demonstrate the ongoing progress in the field of artificial intelligence. With Claude Instant 1.2, Anthropic aims to provide businesses and developers with an improved and safer AI model that can enhance various applications and industries.