frontier model

Definition

Plain language

One of the current most capable AI models — the cutting edge from major labs.

Also called: frontier models

Why it matters: These models define what's currently possible and what new techniques need to beat, so they're the reference point for both capability claims and safety concerns.

For example, when a paper benchmarks against 'frontier models,' it usually means the latest GPT, Claude, and Gemini releases at the time of writing.

Heard on the show

“A hundred and sixty times, researchers asked eight frontier models to hand over the secret PIN sitting in their system prompt.”

Mentioned in 81 episodes

1. 246
   160 Perfect Refusals, And The Refusals Were The Leak
2. 245
   Fifteen Models Ran Football Clubs for Twenty Years, and Size Didn't Decide It
3. 243
   How a Hundred Meaningless Word Choices Add Up to Flip a Model's Answer
4. 242
   Making a Vision Model Better by Showing It Blurry Images
5. 240
   Frontier Models Designed Follow-Ups To Fraudulent Papers 93% Of The Time
6. 238
   How a Cheap Model Reads the Flagship's Secret Reasoning Aloud
7. 236
   Why a Printed 'OPERATOR OVERRIDE' Note Redirects Robot Planners
8. 234
   Two Copies of Gemini Cooperated in a Game Where Betrayal Always Pays
9. 233
   Why a Model Can Grade an Answer But Not Write the Answer Key
10. 232
    Coding Models Can Find the Bad Line, They Just Won't Delete It
11. 217
    Why an AI Called Fourteen Broken Figures Perfect, And What It Reveals About Test-Time Compute
12. 205
    The Same AI, Two Labels: How the Pitch Beat the Product in 162 Sessions
13. 196
    AI Agents Reached Opposite Conclusions From the Same Data — and Passed Review
14. 194
    How a Robot Builds a Debugging Notebook It Can Read, Edit, and Hand to Another Robot
15. 192
    A 32B Open Model Matched Frontier Systems By Learning to Take Notes
16. 191
    How One Researcher Beat GPT-5.2 and Gemini 3 by Judging Their Answers, Not Improving Them
17. 189
    Why Phone Agents Ace the Test and Crash on Your Actual Phone
18. 185
    Aligned to Refuse, Built to Tap: When Phone Agents Know the Task Is a Crime and Do It Anyway
19. 180
    The Bug Where Smart Assistants Read a Fact and Still Forget It
20. 174
    When the AI 'Schemes,' It's Usually Just Lazy or Confused
21. 167
    How Teaching an AI to Predict, Not Act, Made It a Better Actor
22. 166
    A Router That Beats the Frontier Models It Calls
23. 165
    A Free-Lunch Tweak That Lets a Tiny Agent Beat Frontier Giants
24. 160
    Training an AI to Take Its Own Notes, So Its Future Self Works Better
25. 155
    Why a Flawless Demo Makes a Worse Computer-Using Agent, And the Fix
26. 151
    Why More Experience Made This AI Agent Worse, And How to Fix It
27. 147
    Agents Fail at the Body, Not the Brain: A Self-Rewriting Scaffold That Lifts a 9B Model 44 Points
28. 142
    Training a Tiny Model to Run the Plumbing Between an Agent and the World
29. 140
    When a Reasoning Model Says "Let Me Double-Check" After It's Already Decided
30. 139
    When Optimizing One GPU Kernel Quietly Breaks the Whole System
31. 133
    How MiniMax Turned a Reward-Hacking Disaster Into Olympiad Gold
32. 130
    Why AI Agents Coordinate Better Through a Shared Board Than a Boss
33. 128
    How a Model Can Earn Full Reward and Still Resist Training
34. 127
    What Diffusion Language Models Were Missing: A Map, Not an Algorithm
35. 126
    How Coding Agents Can Mine Their Own Failures Into a Self-Targeting Curriculum
36. 125
    AI Coding Agents Run a Marathon, and Fewer Than One in Three Finish
37. 124
    A Cheap Model With the Blueprints Beats Expensive Models Working Blind
38. 119
    Beating Reinforcement Learning Without Ever Touching the Model's Weights
39. 118
    Why the Best-Aligned AI Models Are the Easiest to Trick Into Producing Harm
40. 116
    Why Streaming Half a Reasoning Chain Beats Sending the Whole Thing
41. 112
    When an AI Agent Cheats Without Being Told: Inside the Meta-Agent Challenge
42. 110
    How an Agent Got 44 Points Better by Mining Its Own Scratch Paper
43. 109
    An AI Got Caught Reading the Answer Key, And Why That Catch Matters
44. 105
    The Trojan Is Your Agent's Memory: Why Single-Step Defenses Miss Persistent Attacks
45. 104
    How Making a Research Agent Smarter Quietly Makes It Leak Your Secrets
46. 100
    How a Prompt Wrapper Lets a Frontier Model Play Poker Like an Expert
47. 096
    How Treating an AI Agent's Execution Like Git Recovers a Coordination Penalty
48. 094
    Chain-of-Thought Monitoring Fails Across Languages, and Worst Where It's Needed Most
49. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
50. 087
    When No Agent Reads the Whole Document: A Universal Cliff in Multi-Agent Review
51. 080
    How a Two-Agent Trick Unlocked Large-Scale Training for Computer-Use Agents
52. 079
    An Old Idea From Cognitive Psychology Reshapes How We Reward Reasoning Models
53. 078
    Training a Markdown File: When LLM Self-Improvement Borrows the Discipline of Neural Net Training
54. 069
    When Smarter Models Forecast Worse: The Hidden Failure Mode in LLM Predictions
55. 067
    An AI Just Solved a 1996 Erdős Problem—and the Simplest Agent Won
56. 066
    Why Giving an AI Agent More Tools Can Make It Worse at Using a Computer
57. 065
    One Loop to Optimize Them All: A Universal API for LLM-Driven Discovery
58. 063
    Why Web Agents Are Slow: A Compiler-Style Fix for Computer-Use Latency
59. 062
    Treating Hallucinations as Exploits: A Gate-Based Architecture for Agent Safety
60. 061
    When Helpful Agents Go Sideways: A 404 Error, Campus Security, and Why Alignment Misses This
61. 060
    When Splitting One Model Across Three Agents Doubles Its Accuracy
62. 059
    Firefly's Inversion: Building Verified Tool-Call Training Data by Working Backward
63. 053
    An AI Agent Swapped In Focal Loss And Beat A Human-Tuned Training Script
64. 048
    How a 30B Open Model Reached Olympiad Gold With the Right Recipe
65. 045
    When a Frontier Model Talks Its Own Twin Into Climate Denial
66. 044
    How One Sentence and a Forged History Flip the Most Aligned Models
67. 041
    When the Iteration Teaches the Model to Skip the Iteration
68. 040
    Two Frozen Models Learn to Whisper: Coupling Through Hidden States
69. 039
    When Smarter Agents Get Fooled by Three Extra Nodes in a Database
70. 035
    Why Frontier Agents Ask for Clarification at Exactly the Wrong Moment
71. 031
    When Your AI Assistant Won't Let Go of Old Facts About You
72. 028
    Teaching a Model to Hire Copies of Itself: Recursive Agent Optimization
73. 027
    When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure
74. 024
    An AI Agent That Found 28 Zero-Days in Windows — And What Made It Work
75. 020
    The Compliance Gap: Why AI Says Yes and Does No
76. 017
    When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers
77. 015
    The Audit Number Isn't What You Think: Sycophancy and the Case Against Single-Prompt Bias Tests
78. 008
    Why Long-Horizon AI Agents Get Stuck, and a Milestone-Based Fix That Helps
79. 007
    Exploration Hacking: When Models Sabotage Their Own RL Training
80. 003
    How to Pick the Best of Sixteen Coding Agent Rollouts
81. 001
    When AI Models Quietly Protect Each Other From Shutdown

Related concepts