Prediction Held up AI assessment confidence: 90% certainty 3/5 debate potential 2/5

Kaplan: AI training and inference will see 3x-10x yearly efficiency gains

Jared Kaplan · Scaling and the Road to Human-Level AI | Anthropic Co-founder Jared Kaplan · Y Combinator · Jul 29, 2025 · at 32:02

Anthropic co-founder Jared Kaplan predicts the trajectory of compute efficiency and inference costs for AI systems.

0:00 / 0:26exact quote · 26.5s

▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark

“I think that over time, as AI becomes more and more widespread, I think that we're going to really drive down the cost of inference and training dramatically from where we are right now. Sort of, three X to 10 X gains algorithmically, and in sort of scaling up compute and in inference efficiency per year.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

The cost of AI inference and training will decrease dramatically, achieving 3x to 10x annual efficiency gains across algorithms, compute scaling, and inference.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

Benchmark tracking and empirical studies across the industry confirm that AI training compute efficiency, algorithmic gains, and inference costs have consistently met or exceeded the 3x to 10x annual efficiency gains stated in the claim. Studies by Epoch AI (March 2024 and February 2026) show that pre-training compute needed for a fixed performance level improves by 2.8x to 3x per year from algorithmic innovations alone, with post-training and architectural distillation pushing catch-up training efficiency gains to between 5x and 60x annually. For inference, analyses by the Stanford HAI AI Index (April 2025) and Artificial Analysis (March 2025 and 2026) show that the inference cost for a fixed capability level (e.g., GPT-3.5 and GPT-4 class models) has dropped at annualized rates between 5x and 40x+ (with overall reductions exceeding 280-fold over two years). While total capital expenditure for training frontier models continues to rise due to absolute scale expansion, efficiency per unit of compute and capability consistently achieves Kaplan's projected 3x to 10x annual pace.

Epoch AI research (March 2024 / arXiv:2403.05812) determined that algorithmic efficiency for LLM pre-training halved required compute every 8 months, reflecting a ~2.8x to 3x annual improvement.
Epoch AI and MIRI analyses (December 2025 – February 2026) estimated training compute efficiency gains from distillation and architecture at 5x to 10x+ annually (and up to 16x–60x for catch-up training).
Stanford HAI AI Index Report (April 2025) found LLM inference costs for equivalent capability dropped over 280-fold between November 2022 and October 2024.
Artificial Analysis and Epoch AI joint tracking (March 2025 to September 2026) showed API pricing for GPT-4-class capability fell from ~$30–$60 per million tokens in 2023 to under $0.50 in 2026, representing a 5x to 10x annual price decline.

Checked against

[1] epoch.ai [2] arxiv.org [3] intelligence.org [4] epoch.ai [5] epoch.ai [6] mirantis.com [7] rdworldonline.com [8] valueaddvc.com

Artificial Analysis LLM API price drop trends cost per token reduction AI inference cost reduction rate annual Artificial Analysis Epoch AI algorithmic progress language models compute halving time Stanford AI Index cost of inference training efficiency trends

assessed as of 2026-09-05

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

Mentioned

AI compute efficiency Anthropic Jared Kaplan

⚑ Spot an error? Report it

More from Jared Kaplan

Prediction Not checkable as stated

Kaplan: AI swarms will eventually replicate entire scientific communities

“Eventually we imagine AI models or millions of AI models perhaps working together will be able to do the work That whole human organizations can do. They'll be able to do the kind of work that the entire scientific community currently does.”

Jared Kaplan Jul 29, 2025 ▶ 10:37 Scaling and the Road to Human-Level AI | Anthropic Co-founder Jared Kaplan · Y Combinator

Insight

Kaplan: AI scaling trends are as precise as laws of physics

“We found that there's actually something very, very, very precise and surprising underlying AI training. This really blew us away that there are these nice trends that are as precise as anything that you see in physics or astronomy.”

Jared Kaplan Jul 29, 2025 ▶ 5:15 Scaling and the Road to Human-Level AI | Anthropic Co-founder Jared Kaplan · Y Combinator

Insight

Kaplan: Compute scaling drives AI progress more than researcher cleverness

“Basically you can Scale up the compute in both pre-training and RL and get better and better performance. And I think that's sort of the fundamental thing that is driving AI progress. It's not that AI researchers are really smart or they suddenly got smart. It…”

Jared Kaplan Jul 29, 2025 ▶ 7:54 Scaling and the Road to Human-Level AI | Anthropic Co-founder Jared Kaplan · Y Combinator

Prediction Not checkable as stated

Kaplan: AI may execute multi-month tasks within the next few years

“And this kind of picture suggests that over the next few years, we may reach a point where AI models can do tasks that don't just take us minutes or hours, but days, weeks, months, years, et cetera.”

Jared Kaplan Jul 29, 2025 ▶ 10:23 Scaling and the Road to Human-Level AI | Anthropic Co-founder Jared Kaplan · Y Combinator

Insight

Kaplan: Founders should build products that current AI cannot quite support

“One is I think it's really a good idea to build things that don't quite work yet. This is probably always a good idea. We always want to have ambition, but I think specifically AI models right now are getting better very, very quickly. And I think that's going…”

Jared Kaplan Jul 29, 2025 ▶ 13:48 Scaling and the Road to Human-Level AI | Anthropic Co-founder Jared Kaplan · Y Combinator

Prediction Not checkable as stated

Kaplan: AI scaling curves point smoothly toward human-level AGI

“I think that scaling Really suggests a kind of smooth curve towards what I expect is kind of human level AI or AGI.”

Jared Kaplan Jul 29, 2025 ▶ 17:31 Scaling and the Road to Human-Level AI | Anthropic Co-founder Jared Kaplan · Y Combinator

Kaplan: AI training and inference will see 3x-10x yearly efficiency gains

More from Jared Kaplan

Turn any episode into a week of clips.