Epoch AI

The trillion-dollar question: If AI companies collectively slow the development of AI, will prices also fall more slowly? Read the full report: epoch.ai/publication...

The plunging price of thought epoch.ai

By dividing the latter by the former, we know roughly how much is left. epoch.ai/data/ai-chi... Code for our coverage estimates is available here: colab.research.google.com/drive/1WMhQ...

Google Colab colab.research.google.com

We'll keep auditing and improving the ECI as it evolves. The updated ECI codebase remains open source at github.com/epoch-resea....

GitHub - epoch-research/eci-public: Epoch Capabilities Index github.com

Anthropic says Glasswing has surfaced 10k+ high- or critical-severity vulnerabilities so far (some remain unpublished). OpenAI's Daybreak program likely adds more. The spike in CVEs likely reflects this wave of AI-assisted discovery. Full Data Insight: epoch.ai/data-insigh...

Disclosed CVEs: 3.5× Spike After Claude Mythos | Epoch AI epoch.ai

The AI Chip Users explorer is the next step in our project to map the AI chip supply chain: Which companies sell the most chips: epoch.ai/data/ai-chi... Who bought those chips: epoch.ai/data/ai-chi... The data centers where chips end up: epoch.ai/data/ai-dat...

AI Data Centers epoch.ai

For more details, check out our website! epoch.ai/MirrorCode

MirrorCode: What's the largest software project AI can complete on its own? | Epoch AI | Epoch AI epoch.ai

Compute usage differs from ownership. OpenAI + Anthropic own very little of their hardware, renting compute from cloud providers. Google + Meta parent co's have huge compute fleets, but it's not all allocated to their AI labs. epoch.ai/latest/intr... epoch.ai/data/ai-chi...

Introducing the AI Chip Users Explorer epoch.ai

We’ve started using METR’s Inspect Hawk to run our benchmarks, many of which go into our ECI results. Big thanks to them for open sourcing their great infrastructure, and also for their help getting it set up. Learn more about Hawk here: hawk.metr.org/

Inspect Hawk hawk.metr.org

GPT-5.6 Sol has been climbing Slay the Spire's Ascension ladder on our Twitch channel for a week, no human in the loop. This Thursday, Claude Opus 5 takes over the climb — live, with commentary. Thursday, July 30 · 12:30 PT twitch.tv/epochaiplays

EpochAIPlays - Twitch twitch.tv

Read our full write-up of the results: epoch.ai/publication...

Can AI Spot Mistakes in IKEA Assembly? epoch.ai

Learn more about this Data Insight: epoch.ai/data-insigh...

In August, 25% of math preprints acknowledged AI use, up from 4% in April epoch.ai

Explore the data yourself on our new AI Use in Math Research explorer: epoch.ai/data/arxiv?...

AI Use in Math Research epoch.ai

AI is getting cheaper more quickly than any other transformative tech in history. At a given level of performance, cost has fallen ~47%/quarter since 2023. That’s 4× faster than DNA sequencing, 6× faster than compute, 18× faster than lithium batteries, and (up to 1973) 54× faster than electricity.

Introducing Benchmark Reviews: our new initiative to audit AI benchmarks. We are launching with 15 benchmarks: 4 Verified, 9 Flawed, and 2 with not enough information for a review.

Another problem from FrontierMath: Open Problems has been solved! The solution was elicited by Becker, Greger, and Peters in an interactive session with GPT-6 Astra. Peters originally suggested the problem for the benchmark. He had this to say.

Can AI tell if you've built your IKEA furniture wrong? Our new benchmark, the Furniture Assembly Benchmark (FAB), gives models the manual and a photo of a half-completed piece of furniture and asks them to spot the mistake. The top score has gone from 28% to 80% in just 10 months.

AI appears to be finding software vulnerabilities at scale. In June 2026, 21 notable organizations disclosed ~1,500 high- and critical-severity CVEs, over 3.5× the previous monthly record set before Claude Mythos Preview's release.

How much more AI compute does a dollar buy each year? About 49%, or a doubling every 21 months, based on the chips actually bought each quarter from 2023 through 2025.

Serious cyber vulnerability disclosures keep climbing. In July, 21 major tech organizations published ~2,500 high- and critical-severity CVEs — about 5× the monthly record before Anthropic revealed Claude Mythos Preview could autonomously find software vulnerabilities.

Moonshot's Kimi K3 scores 156 on the Epoch Capabilities Index (ECI), setting a new open-weights record. This places it between Opus 4.6, and GPT 5.4, which released in February and March 2026 respectively, and just ahead of GPT 5.6 Luna.

Introducing EBR-bench, our new benchmark to measure on-the-fly learning. AI repeatedly plays a challenging board game called Earthborne Rangers and tries to learn from its mistakes. So far: no signs of improvement.

Claude Fable 5 scores very well on FrontierMath: Tiers 1–4 (v2), reaching 87% on Tiers 1–3 and 88% on Tier 4. This continues a streak of Anthropic models improving rapidly at math.