Key Takeaways
- Meta and SpaceX both released new models last week that sit cheaper than industry leaders such as OpenAI.
- Businesses are likely to turn to cheaper models as they use more and more AI, which could drive down token prices across the industry.
- When choosing the right AI model, businesses should consider their existing processes and opt for the model best suited for their operations.
Despite being two of the biggest technology companies in the world, Meta and SpaceX have been outpaced by competitors so far in the AI race. However, with the release of new models, both companies are once again throwing their hats into the ring — and, they’re offering cheaper prices.
Recent data suggests these models don’t compare to frontrunners from Anthropic or OpenAI in terms of performance. But, with the amount of US businesses flocking to models like DeepSeek, it’s evident that businesses aren’t against sacrificing slight performance losses for cheaper costs.
Different AI models have different capabilities. When choosing an AI model to scale, businesses should consider which capabilities they need the most in order to supercharge their existing processes.
AI Price Competition Heats Up
Both Meta and SpaceX have announced new AI models in the past week. To mark the release of Grok 4.5 last Wednesday, founder and CEO of SpaceX Elon Musk described the model as offering “frontier intelligence at leading speeds and cost efficiency” on social platform X.
Meta, on the other hand, released Muse Spark 1.1 last Thursday, a model designed specifically for agentic tasks and coding. The model is one of the first from Meta Superintelligence Labs, headed by ex-Scale AI CEO Alexandr Wang.
This just in! View
the top business tech deals for 2026 👨💻
Despite being relatively behind in the AI race, both Grok 4.5 and Muse Spark give SpaceX and Meta the opportunity to play catch up because of their low price offering.
With more businesses adopting AI, many will be keeping an eye on costs, making cheaper models all the more attractive. This, in turn, has the potential to shake up token prices across the industry and potentially lead to a drop.
How Do Meta and SpaceX’s New Models Compare?
Independent AI benchmarking platform Artificial Analysis rates new models Grok 4.5 and Muse Spark 1.1 lower than top competitors Claude Fable 5 and GPT-5.6 Sol. On its Intelligence Index, Grok 4.5 received a 54 score and Muse Spark recorded a 51. While this is slightly lower, it isn’t far off Fable 5 (60) and GPT-5.6 Sol (59), and is higher than models from Google and DeepSeek.
However, Grok 4.5 and Muse Spark are wildly ahead of Claude and GPT when it comes to pricing. Grok 4.5 has a lower weighted average cost per Intelligence Index task, at $0.31 per task, compared to Fable 5 ($2.75) and GPT-5.6 Sol ($1.04). Muse Spark is cheaper than all three, at only $0.26 per task.
Anthropic and OpenAI’s models are better performance-wise according to the Index. However, businesses will be increasingly concerned about pricing as they use more and more AI, and those slight gaps in performance might become negligible if it means they can take advantage of much lower token costs.
Given the popularity of models such as DeepSeek, which costs only $0.04 per task, it’s not out of the question that businesses will start to choose models based on how much they cost. US lawmakers, in fact, have reportedly been looking for ways to curb the use of Chinese models such as DeepSeek by US businesses, because many have looked to the tool as a cheaper alternative to the top players.
So, it may also be the case that businesses using DeepSeek will turn to cheaper models like Grok or Muse Spark if tighter regulation comes into play surrounding the use of Chinese models. If cheaper models start to become more popular as a result, this could have an impact on token pricing across the AI landscape.
How Businesses Should Choose the Right AI Model
Businesses around the world are already spending plenty on AI, as is evident from the tokenmaxxing trend. As more AI is deployed, cost will become a contributing factor to which models are preferred, although businesses should consider a range of factors before choosing.
According to Marina Wyss, Senior Applied Scientist at Twitch: “Businesses tend to default to the newest model from the biggest provider, and that’s rarely the right choice. Each model has a distinct profile of what they’re good at, costs, and what you can use it for. There are three main things to consider.
First, licensing: If you handle customer PII or need to fine-tune on proprietary data, you need to ensure the model you choose supports those use-cases securely.
Second, requirements like task complexity, context length, and how it will be deployed. Finally, make sure to test the finalists on your own data instead of relying solely on benchmarks. You can take 20 to 50 examples from your actual business, run them through two or three candidates, and score for accuracy, cost, and speed.
This will give you much more confidence that you’ve chosen the right model for your business and use case.”
Overall, the model that your business chooses is a significant first step. Ensuring you understand your own needs, and linking these to the tools you’ll need to make your existing processes better, is essential.
