Grok News 03/29/2024 AI Rating: Low

xAI announces Grok-1.5 — 128K-token context and improved reasoning

#Grok#xAI#AI History

xAI has announced a new Grok model, “Grok-1.5.” Its headline feature is an expanded context length of up to 128,000 tokens — about 16 times that of the previous model. Reasoning performance on coding and math also improved significantly.

Details

  • Context length: Supports up to 128,000 tokens, enabling responses that take longer documents and conversation history into account
  • Benchmark results: Scored 50.6% on the MATH benchmark, 90% on GSM8K, and 74.1% on HumanEval
  • Improved reasoning: Emphasized performance gains on coding- and math-related tasks in particular
  • Availability: At the time of announcement, planned for a gradual rollout to X Premium+ subscribers

What happened next

The announcement of Grok-1.5 signaled xAI’s shift from a simple chat-oriented model toward prioritizing long context and reasoning performance. This direction carried through to Grok-2 that August and Grok-3 in February the following year, as xAI rapidly cycled through model generations in a short span of time.