My median estimate for Anthropic's AI R&D value uplift as of mid-October 2026 is approximately 1.5x, with a central 50% interval ranging from 1.4x to 1.8x and an 80% confidence interval ranging from roughly 1.25x to 2.0x. This metric explicitly captures the rate of research output that Anthropic itself judges to be valuable, not raw coding volume or throughput. While raw volume metrics are massive—Claude authored over 80% of merged code by May 2026, and engineers merged 8x as much code per day a
Set against related questions, this baseline estimate was slightly adjusted to ensure it perfectly multiplies with our projected growth ratio to reach the 2029 absolute projection.
The final percentile estimates for Anthropic's AI R&D uplift in value by December 31, 2029, are p10: 1.75x, p25: 2.52x, p50: 4.2x, p75: 8.55x, and p90: 20.0x. This metric explicitly resolves on uplift in value, evaluated against a human-only counterfactual with identical compute, sitting firmly between old-task throughput bounds and unadjusted capability metrics https://metr.org/blog/2026-05-08-task-substitution-and-uplift/.
Curr
Set against related questions, this distribution was slightly adjusted to exactly match the mathematical product of our current baseline estimate and our projected three-year growth ratio .
Estimating the 2026 Baseline (The Denominator) I assess the current AI-assistance multiplier on valuable research output at Anthropic to be roughly 1.6x to 1.8x. Anthropic reports that it does not observe a "sustained AI-attributable 2x acceleration" in overall progress, noting that Opus 5.5 is not substituting for senior researchers [https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf](https://www-cdn.anthropic.com/fc1b44717c85dc068
Set against related questions, this forecast was marginally adjusted to ensure it perfectly bridges the current baseline to the final 2029 absolute projection .
Current State and Baseline We estimate that Anthropic currently holds a modest lead over Google DeepMind in AI R&D uplift in value, placing the present ratio roughly between 1.1 and 1.3. Anthropic has aggressively integrated agentic workflows, with Claude now "leading" (AL4) 26% of its measured AI R&D tasks https://www.anthropic.com/institute/measuring-pace-of-ai-development [https://www.anthropic.com/institute/measuring
Set against related questions, the upper tail of this ratio was slightly compressed because our broader framework heavily weighs structural compute constraints that prevent massive unmitigated divergence.
The median estimate of 0.75 reflects a rapidly shifting landscape where Anthropic's surging enterprise inference volume is closing the gap on OpenAI's structurally larger training and research footprint. While OpenAI has historically dominated total compute, Anthropic's annualized revenue run rate crossed $65B in July 2026, surpassing OpenAI's reported ~$50B run rate techcrunch.com. Because inference dominates total compute at commercial scale, this projected enterprise revenue advantage heavily influ
Accounting for Anthropic's projected enterprise revenue advantage and broader expectations around OpenAI's potential revenue acceleration slightly narrowed the tails of the estimate while leaving the median roughly unchanged.
Anthropic currently holds a higher reported annualized run rate than OpenAI, yielding a headline ratio of approximately 1.45 to 1.5. Spreading Anthropic's reported $16.8B Q3 revenue At Anthropic pre-IPO meet: 'Thinking' caps, surging ... [https://finance.yahoo.com/technology/ai/articles/anthropic-revenue-rises-16-8-132600056.html](https://finance.yahoo.com/technology/ai/articles/anthropic-re
Set against related questions on Anthropic's growth and accounting standards, the ratio estimates were maintained near their previous values, aligning with the projected $112 billion run rate for Anthropic.
On September 8, 2026, METR marked its time-horizons page as "no longer actively updated," warning that measurements above 16 hours are unreliable with its current task suite Task-Completion Time Horizons of Frontier AI Models Task-Completion Time Horizons of Frontier AI Models [https://metr.org/blog/2026-10-14-better-informing-the-public-about-loss-of-control-risks/](https://metr.org/blog/2026-10-14-better-informing-the-public
Set against expectations for the longevity of the current measurement suite, the distribution maintains a strong cluster around the status quo ratio, reflecting a high chance the series remains frozen.
My distribution places the decline in Anthropic's serving cost at a fixed Claude Fable 5 capability level at 9x at the 10th percentile, 20x at the 25th percentile, approximately 45x at the median, 100x at the 75th percentile, and 210x at the 90th percentile. The baseline Fable 5 listed at $50 per million output tokens, scoring 50 on the Artificial Analysis (AA) index artificialanalysis.ai and 66.04% on the Vals Index vals.ai. By late September 2026, just a month after the start date, Claude Sonnet 5.5 ($10
We slightly narrowed the forecast's tails, raising the lower bound to 9x and lowering the upper bound to 210x, to reflect related estimates of baseline hardware costs, recursive R&D scaling, and enterprise AI market dynamics.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited