OpenAI has continued to work on updates for ChatGPT, with a new GPT-4 language model and plugins now in public testing. The company just announced other changes, aimed at making GPT-3 and GPT-4 more useful.
OpenAI has released new versions of its two primary language models — GPT-3.5 Turbo (the default for ChatGPT) and GPT-4 (available in Bing Chat and ChatGPT Pro). The updated models include a new feature called function calling, which makes it easier for software developers to create chatbots that answer questions with external tools. That should improve plugins and other services that use ChatGPT or the official APIs.
The update for GPT-3.5 also includes “more reliable steerability,” which means it should give you more control over its personality in responses. Steerability is what helps prompts like “pretend you’re a pirate” work well, and previously worked much better in GPT-4 than GPT-3.5. There’s also a 32k version of GPT-4 available through the API, which offers “an extended context length for better comprehension of texts.”
OpenAI is also reducing prices for API access. The cost of GPT-3.5 input tokens is now 25% lower, at $0.0015 per 1K input tokens and $0.002 per 1K output tokens. The more powerful 16K version of GPT-3.5 is priced at $0.003 per 1K input tokens and $0.004 per 1K output tokens, and costs for GPT-4 don’t seem to be changing.
Finally, OpenAI is opening up the waitlist for the GPT-4 API to more people. The company is hoping to remove the waitlist entirely soon, which could be a sign that GPT-4 will be available to free ChatGPT users in the not-so-distant future.