Elon Musk: Intelligence is Increasing Exponentially
Elon Musk retweeted a post about the progress of open-source models, simply stating: "Intelligence is increasing exponentially." The post mentioned that GLM 5.3 Flash, DeepSeek V4.1 Flash, Qwen 3.8 Next Flash, and Qwen 3.8 27B have surpassed all models considered cutting-edge as of Christmas 2025 in terms of intelligence and capability, indicating an explosion in open-source development.
Musk did not provide his own evaluation metrics nor specifically name xAI's new weights. This statement follows his earlier assessment that artificial intelligence could surpass the total human intelligence within four to five years, and he has repeatedly moved the timeline for general intelligence closer to the present. He used the term "exponential" to explain that while it appears linear in the short term, it is actually accelerating.
The repost emphasized open-source and small, rapid iterations, rather than maximum training from closed labs. The cutting-edge from ten months ago is now being caught up by cheaper, self-hosted models. Musk uses "exponential" to describe this overall trend of catching up, rather than a single list of rankings.
In the same ecosystem, there are counterexamples: engineers from large companies claim that demand, code, and testing are all generated by programming agents, with humans merely pressing enter. Exponential increases are valid on the model side but do not imply that organizations have fully absorbed them. Musk is focusing on the capability curve, not the on-site workforce.
The information increment is his way of naming the surpassing of open-source over the past ten months as exponential, rather than discussing the ranking of a specific card.
In terms of market mechanisms, this is narrative-driven. Buyers are betting on open-source weights, inference chips, cloud tokens, and AI-related tech stocks; sellers believe that evaluations do not equate to deployable intelligence, and that closed models still control distribution. Capital flows toward training and inference capital expenditures. Beneficiaries are the open-source weight teams and cloud inference; those under pressure are those who view last Christmas's closed frontiers as a moat for pricing.
Source: Public Information
ABAB AI Insight
Musk repeatedly uses the term exponential to describe intelligence, providing a consistent slope for xAI, Tesla robots, and data center capital expenditures. This response targets open-source: ten months later, the once-closed frontiers have been caught up by Flash models and 27B-level models. He does not need to announce new models; he simply recognizes others' catch-up as the curve he has always been drawing.
Capital paths follow computational power. Closed labs lock training with hundreds of billions in capital expenditures; open-source spreads capability to local and cloud inference with downloadable weights. Once the exponential is established, money flows to both ends: larger training clusters and chips and data centers that consume open-source inference. The motivation is to rewrite the timeline from "it's still early" to "it's already in the bend."
In a similar position, it is closer to open-source Linux catching up to proprietary operating systems rather than winning a single benchmark test. The industry phase is that cutting-edge intelligence is spreading from a few APIs to replicable weights. Whoever can productize the capability leap from ten months will rewrite the effective period of "cutting-edge."
Structural changes belong to technological substitution. Last year's largest models have been replaced this year by smaller, rapid iterations of accessible intelligence. The mechanism is that evaluations and open-source releases publicize capability gaps; once the gaps are publicized, the time premium of closed models is compressed; the exponential narrative then writes this compression as inevitable, continuing to fund the next round of clusters.
ABAB News · Laws of Cognition
Exponential looks linear in the short term, but will swallow last year's frontiers in the long term.
Once open-source catches up, the time premium of closed models will shorten.
The faster intelligence rises, the shorter the shelf life of old benchmarks.