Meta Debuts Muse Wonder Full, Ranking Fifth on Vals AI Index and Outperforming OpenAI Models

Meta officially launched its new frontier AI models, Muse Wonder and Muse Wonder Full, on Monday, October 12, 2026, with Muse Wonder Full debuting fifth on the Vals AI Index and outperforming all OpenAI models. Both models feature a one-million-token context window and multimodal input, enhancing long-term user interaction and complex task execution.

Performance Benchmarks and Rankings

The new Muse Wonder models demonstrate significant performance improvements over their predecessor, Muse Spark 1.3. Muse Wonder Full scored 56 points on the Artificial Analysis Intelligence Index v4.3, while Muse Wonder achieved 52 points, an increase from Muse Spark 1.3's 48 points. Notably, Muse Wonder Full secured a position in the top three across six of ten benchmarks and debuted fifth on the overall Vals AI Index, outperforming all OpenAI models on this specific index.

In specialized benchmarks, the models showed substantial gains. On Terminal-Bench 4.0, agentic task performance for the Muse Wonder series improved from 33 percent to 57 percent, with coding tasks showing the most significant enhancement. Muse Wonder Full achieved a score of 68 on the Coding Agents Index, tying with Claude Code with Sonnet 5.5, while Muse Wonder scored 63. For mathematical proofs, Muse Wonder Full solved 94 out of 100 problems on ProofBench, a considerable jump from its predecessor's 58 problems.

Advanced Features: Context Window and Multimodal Input

Both Muse Wonder and Muse Wonder Full models are designed with a one-million-token context window, enabling them to process extensive amounts of information. This large context window is complemented by multimodal input capabilities, allowing the models to accept text, images, video, and audio. A key innovation is the integration of "active memory" and "active focus" features, which are intended to help the models retain user context over extended periods, potentially weeks, and maintain concentration on demanding tasks.

Pricing and Availability

Meta has set the pricing for Muse Wonder at $4 per million input tokens and $20 per million output tokens. This pricing structure is now available following the models' release on October 12, 2026. For more details on pricing across various AI tools, users can visit our AI tools pricing comparison page.

Areas for Improvement

Despite their advancements, the Muse Wonder models exhibit a weak spot in factual knowledge when compared to competitors such as Claude Opus 5.5 and GPT-6 Astra. This indicates an area where future iterations or supplementary knowledge bases may be required to achieve parity with leading models in factual recall and accuracy.

Conclusion

Meta's introduction of Muse Wonder and Muse Wonder Full on October 12, 2026, marks a significant development in the frontier AI landscape. With strong benchmark performances, particularly in agentic tasks and coding, and innovative features like a one-million-token context window and multimodal input, these models position Meta as a key player. While factual knowledge remains an area for development, the initial performance and capabilities suggest a promising trajectory for Meta's AI research and product offerings. Users interested in exploring similar advanced AI capabilities can find more information on top AI tools and latest AI news.

Sources

• Meta releases its first frontier model, Muse Wonder | TechCrunch
• Muse Wonder Puts Meta Back at the AI Frontier | WIRED
• Muse Wonder: A Frontier Model For Personal & Professional AI Agents