30 秒でわかる
Matthew Berman argues that companies can control AI spending by routing different tasks to different models. In this DigitalOcean-sponsored video, he proposes using a more capable model for planning and a cheaper one for code, then describes the sponsor's inference router and a reported customer result.
- 1Rising billsBerman cites reports of large AI bills at Uber and Amazon as motivation, without showing the underlying spending data.
- 2Divide the workHis suggested workflow uses the most capable models to plan and less expensive models to write code.
- 3Sponsored router and exampleDigitalOcean sponsors the video. Berman says its router chooses a model for each prompt and cites a customer's reported 42% reduction in inference costs.
ポイント
- Model routing assigns requests or stages of work to models with different capabilities and prices.
- Berman suggests using a stronger model for planning and a cheaper, faster model for implementation.
- The sponsor reports one customer's 42% inference-cost reduction; the video does not establish that everyone will achieve the same savings or quality.
チャプターごとに整えた全文を読む
The cost of AI
Berman cites reported corporate AI spending to introduce model routing. He argues that routing can reduce costs.
The AI cost reckoning is here: companies are seeing large bills at the end of the month. I cite reports that Uber used its annual AI budget in one quarter and that Amazon spent $500 million in a month. These figures are examples I mention, not spending data established in this short video.
I've been talking about model routing as a way to use a suitable model for each part of a job. My suggestion is to use a top model for planning—figuring out what needs to be done—and delegate the actual code writing to a cheaper, faster model. I argue this can preserve quality at lower cost; the trade-off should be tested for the particular task rather than assumed.
Sponsored inference routing
Berman describes sponsor DigitalOcean's router and a customer example. The savings figure is a reported case, not a general guarantee.
DigitalOcean sponsors this video. I describe its inference router as selecting a model for each prompt, and cite a reported example in which Lava, running more than 130 legal AI agents, reduced inference costs by 42% after enabling the router without code changes. That's one customer claim, not a promise of the same result for every workload.
同じチャンネルの記事
34:21EN
Matthew Berman· 2026年10月27日Matthew Berman Examines Anthropic’s IPO Filing, AI Growth and GovernanceMatthew Berman, Sam Altman · Forward FutureThe presenter reviews the S-1’s financial metrics, compute strategy and voting structure while questioning the tension between rapid AI development and safety.
0:26ENMatthew Berman· 2026年10月25日Sam Altman on Human Values and Problem-Solving in an AI WorldSam Altman · OpenAIIn a short interview excerpt, Altman considers the enjoyment of solving hard problems and the human role in interpreting results.
0:36ENMatthew Berman· 2026年10月25日Kimi K3.1 and a Rubik’s Cube Solver DemonstrationMatthew Berman · Forward FutureMatthew Berman shows a scrambled virtual cube being solved and gives the Kimi K3.1 test a favorable assessment.
0:21ENMatthew Berman· 2026年10月24日An AI-Generated Kimi Benchmark Chart Used Invented ScoresMatthew Berman · Forward FutureMatthew Berman asks for a graph of Kimi K3.1 Pro benchmarks, then discovers that the displayed figures are illustrative placeholders.
25:50ENMatthew Berman· 2026年10月24日Kimi K3.1 Pro Tested: Strong 3D Demos, Uneven Results and Competitive PricingMatthew Berman · Forward FutureMatthew Berman tests the open-weight model on coding, 3D rendering and visual explanations, then compares benchmark results and prices.
32:20ENMatthew Berman· 2026年10月23日Who Is Local AI For? The Costs, Trade-offs and a Hybrid FutureMatthew Berman · Forward FutureMatthew Berman weighs data control and independence against hardware cost, setup complexity and inference speed.