Beyond the Big Three: Grok 4.7, DeepSeek V4.1-Flash and Meta's Muse Spark 1.3

xAI: Grok 4.7 (September 21)

xAI released Grok 4.7 on September 21, 2026, and calls it its most capable model for coding and knowledge work. It follows Grok 4.5 (July 8) and Grok 4.6 (August 12).

According to launch coverage:

  • It was trained for harder problems that can take hours, using a larger base model and a longer reinforcement-learning run.
  • It has the same price as Grok 4.6: $2 per million input tokens and $6 per million output tokens, with a 500,000-token context window.
  • It launched alongside Grok Build, xAI's terminal-based coding agent, and is available in Cursor and the Grok API.
  • Reported scores: CursorBench 4.0 46.3% (Grok 4.6: 40.4%).

Starting with Grok 4.5, the models are co-developed with Cursor, which is described as an incoming subsidiary of SpaceXAI. In July, a researcher reported that Grok Build had uploaded users' entire source repositories, including API keys and files users had excluded, to a SpaceXAI cloud storage bucket.

DeepSeek: V4.1-Flash (September 10)

DeepSeek released DeepSeek-V4.1-Flash on September 10, 2026. It is the smallest model in a new architecture family.

  • Size: a 552-billion-parameter mixture-of-experts model with 8B parameters active per token on input and 16B on output, using a new "Causal Encoder-Decoder" architecture.
  • Performance claim: DeepSeek says multiple parties' tests put it ahead of its much larger flagship, V4-Pro, on performance, cost, speed and total runtime.
  • Open weights: yes. SiliconANGLE reports the weights are MIT-licensed.
  • API change: from September 14, requests to V4-Pro are answered by V4.1-Flash and billed at Flash rates until a V4.1-Pro launches.

Meta: Muse Spark 1.3 (September 2)

Meta Superintelligence Labs released Muse Spark 1.3 on September 2, 2026, with paid developer access. Bloomberg reported that the update brings Meta closer to OpenAI and Anthropic. Muse Spark is proprietary and handles about a million tokens of context.

For people running models locally, Meta's more relevant release came on August 10: Muse Glimmer, a 30-billion-parameter open-weight model under the Apache 2.0 license that accepts both text and images. Meta says it can run entirely offline on a single 24 GB consumer GPU.

Sources

This article was drafted with AI assistance and checked against the sources above. Performance figures are the vendors' own.