--- schema: 1 type: video title: >- Scaling the Next Paradigm of Heterogeneous Intelligence — Adrian Bertagnoli, Callosum description: >- A mixture of Qwen 3 VL8B and Kimi K2.5 beat the state of the art on Video Web Arena, outperforming the leading GPT and Gemini models by 18 and 25 percent while costing 3.7 times less and running 3 times faster. The reason it worked is that visual web navigation decomposes into subtasks that do not all need a frontier model: routing zoom and visual parsing to a smaller model alone produced 11x speed and 43x cost improvements on those steps. Adrian Bertagnoli from Callosum makes the case that the GPU cluster era of identical hardware and monolithic models is ending. Heterogeneous intelligence treats model architectures, chip types, and workflows as variables to optimize together. A second result: running recursive long context reasoning tasks on Cerebras instead of a frontier model cuts cost by 7x and latency by 5x while matching accuracy. Callosum is building the automation layer that routes tasks to the right chip and model without bespoke decisions for each subtask. Speaker info: - https://www.linkedin.com/in/adrian-bertagnoli-bb3467178/ Timestamps 0:14 Introduction and definition of heterogeneous intelligence 0:56 Limitations of the current homogeneous intelligence paradigm 1:36 Evolution toward mild heterogeneity (MoE, multi-agent systems, hardware disaggregation) 3:24 The rationale for heterogeneity: complexity and multi-step problem solving 4:26 Mathematical formalization of the production function and skill distribution 5:56 Practical implementation of heterogeneous workflows 6:55 Case study: Recursive language models and context management 9:05 Results on Ulong benchmarks (Cerebras/Sambanova performance) 10:20 Case study: Visual web navigation and Video Web Arena performance 12:02 Offloading subtasks to smaller models for speed and cost efficiency 12:38 The future of compute: Moving to a heterogeneous, multi-agent stack 13:10 Partnership with the UK's Arya institute 13:31 Closing summary and outlook on hardware/software co-evolution 14:01 Q&A: Automation layer for task routing --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/WRBNDpUhsJQ-scaling-the-next-paradigm-of-heterogeneous-intelligence Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- canonical: >- https://liquidgold.cc/channel/aidotengineer/explore/video/scaling-the-next-paradigm-of-heterogeneous-intellig-e639a766ce2f markdown: >- https://liquidgold.cc/channel/aidotengineer/video/scaling-the-next-paradigm-of-heterogeneous-intellig-e639a766ce2f.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T13:14:25.939Z' language: null children: [] videoId: WRBNDpUhsJQ durationSeconds: 913 publishedAt: '2026-05-24T14:00:06.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 25 endSeconds: 163 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=25' heldReason: rights - ordinal: 1 text: null startSeconds: 25 endSeconds: 163 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=25' heldReason: rights - ordinal: 2 text: null startSeconds: 173 endSeconds: 261 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=173' heldReason: rights - ordinal: 3 text: null startSeconds: 173 endSeconds: 261 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=173' heldReason: rights - ordinal: 4 text: null startSeconds: 272 endSeconds: 354 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=272' heldReason: rights - ordinal: 5 text: null startSeconds: 272 endSeconds: 354 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=272' heldReason: rights - ordinal: 6 text: null startSeconds: 364 endSeconds: 452 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=364' heldReason: rights - ordinal: 7 text: null startSeconds: 364 endSeconds: 452 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=364' heldReason: rights - ordinal: 8 text: null startSeconds: 463 endSeconds: 560 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=463' heldReason: rights - ordinal: 9 text: null startSeconds: 463 endSeconds: 560 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=463' heldReason: rights - ordinal: 10 text: null startSeconds: 560 endSeconds: 571 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=560' heldReason: rights - ordinal: 11 text: null startSeconds: 571 endSeconds: 697 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=571' heldReason: rights - ordinal: 12 text: null startSeconds: 571 endSeconds: 697 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=571' heldReason: rights - ordinal: 13 text: null startSeconds: 571 endSeconds: 697 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=571' heldReason: rights - ordinal: 14 text: null startSeconds: 707 endSeconds: 819 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=707' heldReason: rights - ordinal: 15 text: null startSeconds: 707 endSeconds: 819 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=707' heldReason: rights - ordinal: 16 text: null startSeconds: 707 endSeconds: 819 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=707' heldReason: rights - ordinal: 17 text: null startSeconds: 829 endSeconds: 913 watch: 'https://liquidgold.cc/watch/WRBNDpUhsJQ?t=829' heldReason: rights --- # Scaling the Next Paradigm of Heterogeneous Intelligence — Adrian Bertagnoli, Callosum A mixture of Qwen 3 VL8B and Kimi K2.5 beat the state of the art on Video Web Arena, outperforming the leading GPT and Gemini models by 18 and 25 percent while costing 3.7 times less and running 3 times faster. The reason it worked is that visual web navigation decomposes into subtasks that do not all need a frontier model: routing zoom and visual parsing to a smaller model alone produced 11x speed and 43x cost improvements on those steps. Adrian Bertagnoli from Callosum makes the case that the GPU cluster era of identical hardware and monolithic models is ending. Heterogeneous intelligence treats model architectures, chip types, and workflows as variables to optimize together. A second result: running recursive long context reasoning tasks on Cerebras instead of a frontier model cuts cost by 7x and latency by 5x while matching accuracy. Callosum is building the automation layer that routes tasks to the right chip and model without bespoke decisions for each subtask. Speaker info: - https://www.linkedin.com/in/adrian-bertagnoli-bb3467178/ Timestamps 0:14 Introduction and definition of heterogeneous intelligence 0:56 Limitations of the current homogeneous intelligence paradigm 1:36 Evolution toward mild heterogeneity (MoE, multi-agent systems, hardware disaggregation) 3:24 The rationale for heterogeneity: complexity and multi-step problem solving 4:26 Mathematical formalization of the production function and skill distribution 5:56 Practical implementation of heterogeneous workflows 6:55 Case study: Recursive language models and context management 9:05 Results on Ulong benchmarks (Cerebras/Sambanova performance) 10:20 Case study: Visual web navigation and Video Web Arena performance 12:02 Offloading subtasks to smaller models for speed and cost efficiency 12:38 The future of compute: Moving to a heterogeneous, multi-agent stack 13:10 Partnership with the UK's Arya institute 13:31 Closing summary and outlook on hardware/software co-evolution 14:01 Q\&A: Automation layer for task routing --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/WRBNDpUhsJQ-scaling-the-next-paradigm-of-heterogeneous-intelligence Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/scaling-the-next-paradigm-of-heterogeneous-intellig-e639a766ce2f · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks --- #### Chunk 0: Document excerpt (0:14 to 0:25) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 1: Document excerpt (0:25 to 2:43) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 2: Document excerpt (2:43 to 4:32) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 3: Document excerpt (4:32 to 5:54) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 4: Document excerpt (5:54 to 7:32) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 5: Document excerpt (7:32 to 9:20) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 6: Document excerpt (9:20 to 9:31) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 7: Document excerpt (9:31 to 11:37) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 8: Document excerpt (11:37 to 13:39) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 9: Document excerpt (13:39 to 15:13) > Held: transcript rights do not permit this chunk's text.