Anthropic, OpenAI Launch New AI Models Amid Calls for a Slowdown

Anthropic and OpenAI, two leading players in the artificial intelligence (AI) industry, have unveiled advanced new AI models, Frontier AI, as their latest innovations. However, despite this announcement, the companies' leaders have advocated for a slowdown in AI development, raising concerns about the pace of technological progress.

OpenAI is currently developing successors to Fable 5.1 and GPT-6 Astra. As reported by Engadget, these new models provide similar performance at a lower cost compared to their predecessors. On the other hand, Anthropic has introduced Opus 5.5, its most advanced AI model to date. Anthropic claims that Opus 5.5 outperforms other models in handling complex tasks, including identifying and fixing software inefficiencies, as well as financial analysis and business operations.

Opus 5.5 is specifically targeted towards Anthropic's corporate clients. Benchmarks conducted by Anthropic suggest that Opus 5.5 surpasses OpenAI's GPT-6 Astra in agentic coding tasks on Terminal-Bench 4.0 and FrontierCode v1.1 (Main), offering a more advantageous proposition for developers. Beyond the tasks it can handle, Anthropic highlights that Opus 5.5 produces writing that is easier to understand and violates constraints about 85 percent less often than its predecessor, indicating a reduced likelihood of violating directives.

Additionally, Anthropic asserts that Opus 5.5 has a lower cost per token, with input tokens priced at US$4 (Rp71,000) and output tokens at US$20 (Rp350,000), which is more affordable than the token pricing for Opus 5 (US$5 (Rp89,000) for input tokens and US$25 (Rp445,000) for output tokens).

OpenAI's new models, GPT-6 Sol and GPT-6 Luna, emphasize efficiency. Both models use training methods similar to those employed for GPT-6 Astra. These enhancements encompass various aspects, such as professional work, programming, and computer usage. Notably, OpenAI offers these models at a promotional price that is up to 50 percent lower than the initial pricing for GPT-5.6.

For instance, GPT-6 Sol input tokens cost US$2 (Rp35,000), while output tokens are priced at US$10 (Rp178,000). In comparison, GPT-6 Luna input tokens are priced at US$0.10 (Rp1,700), and output tokens at US$0.50 (Rp8,900).

In terms of improvements, OpenAI claims that the new GPT-6 models are better at delivering accurate information, with Sol making approximately half as many errors as its predecessor. Regarding coding, OpenAI states that GPT-6 Sol can match the performance of Fable 5.1 at a lower cost. Furthermore, these models communicate more clearly and benefit from improvements in prompt caching, allowing them to reuse more context and respond more swiftly.

OpenAI asserts that the GPT-6 model demonstrates greater alignment with company guidelines, based on internal testing. For example, it is less likely to fabricate results of coding work and more inclined to reject attempts to circumvent security measures when provided with unsafe instructions.