Predibase vs Improbable

Neither company has publicly disclosed a valuation, making a definitive size comparison difficult. Predibase employs 50-200 people, while Improbable has 500-1000 employees.

Improbable has raised more in total funding at $700M, compared to Predibase's $16M — a gap of $684M. Combined, the two companies have completed 4 known funding rounds.

Improbable leads with an Awaira Score of 68/100, while Predibase sits at 55/100. That 13-point gap reflects real differences in funding, scale, and traction — it's not a vanity metric.

Predibase was founded by Travis Addair in 2021. Improbable was founded by Herman Narula in 2012. Visit each company's profile on Awaira for a full founder biography.

Predibase: Predibase provides a fine-tuning and deployment platform purpose-built for large language models, enabling engineering teams to adapt open-source foundation models to their specific domains and production requirements without building the supporting infrastructure from scratch. The platform includes LoRA-based fine-tuning, evaluation pipelines, and a dedicated serving layer optimized for fine-tuned model latency.\n\nThe company raised approximately 16 million USD and serves teams at enterprises and growth-stage companies that have concluded that generic LLM APIs are insufficient for their accuracy requirements and want domain-adapted models they own rather than rent. Predibase is built on top of Ludwig, an open-source declarative ML framework developed at Uber and maintained by the same team.\n\nThe fine-tuning market is growing as production AI teams discover that base models underperform on specialized tasks and that instruction fine-tuning on domain-specific data meaningfully improves output quality. Predibase competes with Modal, Anyscale, and cloud-native fine-tuning services, but its focus on the complete workflow from training through serving rather than raw compute rental positions it closer to a managed ML platform than an infrastructure provider. Improbable: Improbable develops large-scale simulation technology and virtual world infrastructure, originally focused on cloud-distributed game server simulation and subsequently pivoting to AI-powered synthetic environment generation and metaverse platform development. The London company built SpatialOS, a cloud platform for running distributed simulations of large, persistent virtual worlds, and has applied simulation capabilities to defence, urban planning, and entertainment applications.\n\nThe company raised approximately $700 million including a landmark $502 million SoftBank round in 2017 that was one of the largest venture rounds in European technology history. Improbable has undergone multiple strategic pivots, moving from gaming infrastructure to defence simulation contracts with clients including the UK Ministry of Defence, which uses simulation environments for training and wargaming. The company also built MSquared, a metaverse interoperability network, before refocusing on defence and AI simulation work.\n\nImprobable operates in an AI simulation market where demand from defence agencies, autonomous vehicle companies, and robotics firms has grown substantially. The company competes with simulation platforms from Epic Games (Unreal Engine), NVIDIA (Omniverse), and specialist defence simulation vendors. Its SoftBank backing and large cash reserves have enabled it to survive multiple market pivot cycles that would have ended less well-funded companies, and its current focus on AI-powered synthetic data generation for defence represents a growing government budget category.

Improbable got there first, launching in 2012 — that's 9 years of extra runway. Predibase didn't arrive until 2021. In AI, that kind of head start means more training data, deeper customer relationships, and a bigger talent moat.

Predibase has about 50-200 employees; Improbable has about 500-1000. A bigger team usually means more revenue or heavier VC backing, but in AI, small teams can build at massive scale.

Yes — they're direct rivals. Both Predibase and Improbable compete in ML Platform, targeting many of the same buyers. If you're evaluating one, you should be looking at the other.