Anthropic announced on August 9, 2023 the release of Claude Instant 1.2, a faster and lower-priced variant of its Claude family of models, available through the Anthropic API.
Capabilities of Instant 1.2
Claude Instant 1.2 is positioned as a practical, cost-effective model suitable for casual dialogue, text analysis, summarization, and document comprehension. The update incorporates strengths from Claude 2 into real-world use cases.
Performance improvements: math, coding, reasoning
According to Anthropic, Instant 1.2 shows significant gains in several key areas, notably mathematical problem solving, coding, and general reasoning. Reported benchmark results include:
- Codex evaluation: Instant 1.2 scores 58.7%, versus 52.8% for Claude Instant 1.1.
- GSM8K (math benchmark): Instant 1.2 scores 86.7%, compared to 80.9% for Claude Instant 1.1.
The model produces longer, more structured responses and follows formatting instructions more faithfully. Anthropic also reports improvements in quote extraction, multilingual capabilities, and question answering.
Safety and reliability
Anthropic states that Instant 1.2 has improved safety characteristics: it hallucinates less and is more resistant to jailbreak attacks. These improvements are supported by the company’s automated red-teaming evaluation. In their safety metric, lower scores indicate better performance.
Access and business options
Developers and businesses can call Claude Instant 1.2 via the Anthropic API; pricing details are available on Anthropic’s website. Companies interested in partnering with Anthropic can register their interest through the company’s business channels.
Summary
Claude Instant 1.2 is presented as a faster, lower-cost option within the Claude family that transfers advances from Claude 2 into practical applications, achieves higher benchmark scores in math and coding, and shows measurable safety and capability improvements.
