Theme

The frontier is spiky

Explore this theme’s evidence over time →

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to December 12, 2025 (theinnermostloop.substack.com)

No single model leads everywhere. Capability arrives unevenly across domains, which is why the corpus tracks per-benchmark results rather than a single ranking.

Development timeline

Newest reporting dates first. Reporting dates may differ from event dates.

2026-10-03: Argon reportedly leads Text Arena

2026-10-03: Argon reportedly matches Astra at lower cost

2026-10-03: Opus 5.5 reportedly leads Epoch’s index

2026-10-03: GPT-6.1 Sol reportedly leads MathArena

2026-09-17: Aaronson cites formal mathematics as evidence of a transition

2026-08-12: A model reaches the human baseline on a benchmark built to be impossible

2026-08-04: A benchmark tests rebuilding whole projects from scratch and passing every test

2026-07-26: A model quadruples the best score on the hardest reasoning benchmark

2026-07-18: Four launches in eight days lift six labs above an intelligence threshold

2026-07-11: A 14B model beats every frontier giant at ideological evenhandedness

2026-07-11: A challenger ties for first in frontend coding at half the price

2026-07-11: A lab delays its flagship a second time to retrain a fresh base

2026-07-08: A rival's single turn is said to do the work of ten

2026-07-04: A throttled model scores ten points below itself and still leads

2026-07-03: Seventeen models have taken the lead, each reigning about seven weeks

2026-07-02: Agents graded on taste solve under a quarter of senior-level tasks

2026-07-01: A lab wires sixty-plus databases into one reproducible science workbench

2026-06-28: A company pairs two frontier models by comparative advantage

2026-06-22: Frontier models flail at long-horizon strategy in a video game

2026-06-21: A biology model raises the pass rate on 750 expert tasks

2026-06-20: An AlphaFold Nobel laureate leaves for a rival lab

2026-06-19: A model runs robodog tasks twenty times faster than the best humans

2026-06-15: An open model beats the frontier on ML engineering under cost limits

2026-06-14: The intelligence frontier chart moves backward for the first time

2026-06-10: One model tops effectively every board at once

2026-06-07: Frontier agents resolve under a fifth of a multi-hour task marathon

2026-06-02: The best model reaches 1.5% of human efficiency on ARC-AGI-3

2026-05-29: The safest frontier model is also the strongest

2026-05-20: A fast model beats its own flagship while the flagship slips

2026-05-17: A shipped model is reported improving every few days

2026-05-16: Rank by exploitation and one model leads; rank by economic value and it reverses

2026-05-08: A benchmark asks agents to rebuild whole codebases from a binary

2026-05-03: The hardest benchmark starts to move

2026-05-01: An early checkpoint matches an unreleased rival on cyber tasks

2026-04-30: A probe puts the frontier model near ten trillion parameters

2026-04-27: A frontier model's service life falls from 21 months to 49 days

2026-04-20: AGI is demoted from prophecy to a point on a release calendar

2026-04-17: A point release lands midway to a model the public cannot use

2026-04-09: The first model class trained at scale on Blackwell arrives mid-handoff

2026-04-09: One cluster runs seven simultaneous training runs

2026-04-09: A model crushes data-quality benchmarks and flubs reasoning ones

2026-04-08: Meta Superintelligence Labs ships its first model

2026-04-08: One model sweeps the board and jumps the capability curve

2026-03-29: Three frontier releases land in one month

2026-03-27: A new benchmark returns frontier models to near zero

2026-03-27: A tier above Opus leaks and is deleted

2026-03-18: A distilled model approaches the full one at twice the speed

2026-03-13: Meta discusses licensing a rival's model to power its own products

2026-03-04: A model trims hallucinations 30% and dials back moralizing

2026-03-04: A specialist model more than doubles the generalist on hard health questions

2026-02-24: A model passes the car wash test

2026-02-13: Seven people on Earth can still beat it at competitive programming

2026-02-07: Opus 4.6 takes the top index spot as markets price the next month

2026-02-06: Opus 4.6 sets a new high on Humanity's Last Exam

2026-02-06: A benchmark record is beaten under thirty minutes later

2026-01-24: A 10B Chinese model claims to beat models twenty times its size

2026-01-10: Grok 5 is a seven-trillion-parameter model

2025-12-30: The Remote Labor Index tops out at 2.5%

2025-12-18: Gemini 3 Flash posts elite scores across four benchmarks

2025-12-12: Benchmarks show distinct per-model strengths

Knowledge graphopen in browser →

Cite or correct this record

Citation formatCitation text

The frontier is spiky
Innermost Loop KB — secondary record
Reporting date: unknown
KB permalink: https://nicholsn.github.io/innermost-loop-kb/themes/spiky-frontier
Attributed sources:
- Welcome to December 12, 2025 — https://theinnermostloop.substack.com/p/welcome-to-december-12-2025
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.