The market for large language models (LLMs) remains in motion—faster, denser, and more diverse than ever. With the Insiders LLM Bench­mar­king for Q4 2025, we once again provide clarity in an envi­ron­ment where new models emerge every month and existing variants continue to be refined.

For this edition, we nearly doubled the dataset and made the documents signi­fi­cantly more complex. This allows the bench­mar­king to reflect real pro­duc­tive IDP workflows even more accu­ra­tely — although the higher dif­fi­culty level slightly lowers the average scores.

A REALISTIC COM­PA­RISON UNDER TOUGHER CON­DI­TIONS

The current bench­mar­king covers 24 models, including new entrants such as Claude 4.5 Sonnet, Gemini 3 Pro, and GPT‑5.1. Models whose suc­ces­sors now offer com­pa­rable per­for­mance at similar cost were removed.

Once again, dedicated reasoning models deliver strong results in clas­si­fi­ca­tion and extra­c­tion. At the same time, the same struc­tural drawbacks seen in the previous benchmark reappear: longer pro­ces­sing times, higher token costs, and less pre­dic­table operation in pro­duc­tion. For example, GPT‑5 and GPT‑4.1 achieve excellent overall per­for­mance scores of 87.3 and 84.7, respec­tively — but come with notable dis­ad­van­tages when it comes to data pro­tec­tion or pro­ces­sing speed.

Compared to last quarter, the number of EU-hosted models in our selection has increased — though they remain scarce in the overall market.

SPE­CIA­LIZA­TION MAKES THE REAL DIF­FE­RENCE

Our own model again shows the strongest progress: OvAItion Private LLM improves by more than two per­cen­tage points despite the more demanding test data and, for the first time, approa­ches well-known models like Claude 4.5 Haiku. This is no coin­ci­dence — our current Private LLM is being merged with the announced OvAItion LLM to form the “OvAItion Private LLM,” combining maximum security with steadily improving quality and spe­cia­liza­tion for the IDP envi­ron­ment of our customers and partners.

The takeaway is clear: spe­cia­liza­tion beats size. While large foun­da­tion models make only incre­mental advances, domain-specific models deliver the meaningful quality gains.

DATA SOVE­REIGNTY AS A STRATEGIC ADVANTAGE

In regulated envi­ron­ments in par­ti­cular, operating a self-hosted LLM is becoming incre­asingly important. Orga­niza­tions benefit from full data control, C5-certified security, pre­dic­table costs, and maximum adap­ta­bi­lity. The trend is rein­forced: high per­for­mance and regu­la­tory com­pli­ance rarely coexist in global models — but are achie­vable in private deploy­ments.

KEY INSIGHTS FROM THE Q4 BENCHMARK

• Large foun­da­tion models operate at a high level, but progress slows noti­ce­ably in the IDP context

• Reasoning models achieve strong scores but are often inef­fi­cient in practice

• Under real IDP con­di­tions, benefits remain limited: overhead outweighs added quality

• High per­for­mance and regu­la­tory security seldom go hand in hand

BEST-OF-BREED AS A LONG-TERM STRATEGY

Insiders con­sis­t­ently pursues a best-of-breed approach: we con­ti­nuously test all relevant models, integrate them through the OvAItion Engine, and enable customers to flexibly use exactly the models that best meet their requi­re­ments. Com­ple­men­tary mecha­nisms such as Green Voting auto­ma­ti­cally safeguard result quality and reduce manual post-pro­ces­sing.

This keeps the Insiders LLM Bench­mar­king a reliable point of ori­en­ta­tion in a market that evolves faster than any single provider can keep up with.

For indi­vi­dual bench­mar­kings, our AI experts are happy to advise you per­so­nally: