Meta Releases The Most Powerful Model Ever, Muse Spark 1.3, With Encoding To Compete With GPT-5.6, Closing In On The AI Race To Top Players!
Meta Platforms has released its most powerful AI model to date, Muse Spark 1.3. Meta's Chief AI Officer stated that this is the company's biggest leap in model performance so far, especially in coding and agent tasks, where Meta has further closed in on top competitors like OpenAI and Anthropic. Starting Wednesday, developers can access and pay Muse Spark 1.3 via the Meta Model API. Meta also stated that this update will soon be rolled out to users of social media products such as Instagram, Facebook, and Meta AI. However, it should be noted that direct comparisons between models are not straightforward. Different models have their own strengths and weaknesses on various tasks, and benchmark parameters may be deliberately optimized to not fully reflect real use cases.
According to Meta's published benchmark data, Muse Spark 1.3 scored 75.4 in DeepSWE, higher than GPT-5.6 Sol and Opus 5. However, Meta acknowledged that competitors' models still maintained their lead across multiple benchmarks. Muse Spark 1.3 showed significant improvements in efficiency. The new model required about 20% fewer tool calls and reduced token consumption by about 25%. This means the model takes fewer detours in daily workloads, responds more concisely, and reduces unnecessary interaction rounds.
This efficiency improvement is especially important for coding tasks. Long tasks require the model to continuously remember the user's initial needs and execute them stably based on them. Muse Spark 1.3 can handle multiple workflows simultaneously in the same conversation without opening multiple separate sessions; It also better manages long and complex instructions, preserving key details between multiple tasks. Meta also states that the new model is more aware of its own limitations and is more willing to proactively ask users for clarification when requests are ambiguous, rather than continuing with erroneous assumptions. Before irreversible operations are involved, the model also seeks confirmation to reduce the risk of misoperations.
On the commercialization side, Meta chose to keep the API price for Muse Spark 1.3 unchanged. Standard pricing is $1.25 per million input tokens, $0.15 per million cached input tokens, and $4.25 per million output tokens. As with the previous version, Meta continues to offer a cheaper "contributor" tier to developers who allow their data to improve models, priced at $0.10 per million input tokens and $0.20 per million output tokens. This is Meta's ongoing attempt at AI commercialization. In July this year, Meta began charging developers for Muse Spark 1.1 for the first time.
Market Insight:
Currently, Meta is investing hundreds of billions of dollars, trying to catch up with leaders in the rapidly changing AI race. However, Meta's large-scale spending has also drawn investor attention, with the market hoping for clearer investment returns. Against this backdrop, Meta has begun to expand into cloud infrastructure, selling AI computing power and model access rights.
