SWE-Bench

Measuring Exponential Trends Rising (in AI) — Joel Becker, METR

Feb 27, 2026 · 1:05:12

Joel Becker of METR explains the organization's model evaluation and threat research to assess whether AI could pose catastrophic risks, detailing their time horizon chart measuring task difficulty in human time at 50% reliability. He describes how tasks are selected for economic relevance and auto-gradability, and why time horizon is often misinterpreted as agent runtime. The episode covers Opus 4.5's surprising jump, challenges redoing developer productivity RCTs as workflows change, and why current models aren't yet catastrophically dangerous. Becker discusses potential capability explosions if R&D loops fully automate, links between compute growth slowdowns and slower capability progress, and his Manifold trading story driven by a charity market he could influence. He previews METR's 2026 plans for monitoring and risk assessment, and their hiring.

The End of SWE-Bench Verified — Mia Glaese & Olivia Watkins, OpenAI Frontier Evals

Feb 23, 2026 · 27:10

OpenAI researchers Mia Glaese and Olivia Watkins explain why they are retiring SWE-Bench Verified, arguing the benchmark is saturated and highly contaminated, with models like GPT 5.2 regurgitating ground-truth solutions or task IDs. They detail how a contamination auditor agent found evidence across Claude Opus 4.5 and Gemini Flash, and how over half of unsolved problems had unfair tests (e.g., requiring specific function names not in the spec). The team is pivoting to SWE-Bench Pro from Scale, which features harder, longer tasks (1-4+ hours), more diverse repos and languages, and far less contamination. They advocate for future coding evals to measure open-ended design decisions, code quality, maintainability, and real-world product building, while also calling for more real-world usage metrics to track AI's impact on jobs and productivity.

[State of Code Evals] After SWE-bench, Code Clash & SOTA Coding Benchmarks recap — John Yang

Dec 31, 2025 · 17:45

John Yang, creator of SWE-bench, explains the benchmark's evolution from an ignored project to the industry standard after Cognition's Devin launch, and introduces CodeClash as a new paradigm for evaluating long-horizon coding agents via programming tournaments. Yang covers SWE-bench's expansion into nine languages (JavaScript, Rust, Java, C, Ruby) and multimodal tasks, the proliferation of independent variants like SWE-bench Pro and SWE-bench Live, and the shift from unit tests to iterative agent competitions. He addresses the Tau-bench controversy over impossible tasks, arguing they serve as cheat detection, and weighs the tension between long autonomy (5-hour runs) and interactivity (Cognition's fast back-and-forth). Yang highlights the academic data problem—companies like Cognition have rich user interaction data, while academics need user simulators or compelling products like LMArena—and positions CodeClash as a testbed for studying human-AI collaboration by freezing model capability and varying interaction setups.

Terminal-Bench: Pushing Claude Code, OpenAI Codex, Factory Droid, et al to the limits

Oct 18, 2025 · 35:27

Alex Shaw and Mike Merrill, creators of Terminal-Bench, explain how their coding agent benchmark became an industry standard adopted by Anthropic, OpenAI, and leading agent companies. They bet on terminal-based interaction over GUI because text is the modality that works best with models, and designed tasks as containerized environments with instructions and test scripts. Nicholas Carlini at Anthropic became an early champion, leading to Terminal-Bench being featured on Claude’s model card without notice. They built the minimal Terminus agent to isolate model capabilities from agent optimizations, and their framework adapts existing benchmarks like SWE-bench. Future plans include cloud hosting, multi-dimensional evaluation incorporating cost and economic value, and enabling RL post-training on benchmark tasks.

ChatGPT Codex: The Missing Manual

May 16, 2025 · 53:32

Josh Ma and Alexander Embiricos from OpenAI introduce ChatGPT Codex, a cloud-hosted autonomous software engineer that runs in its own sandbox, arguing it delivers value through single-shot, long-horizon task completion. They explain that Codex evolved from internal experiments (codenamed WHAM) and now offers a 60-tasks-per-hour rate limit with a one-hour hard cutoff, emphasizing it is not meant for interactive use but for delegation—users fire off tasks and go about their day. Best practices include using agents.md, installing linters and formatters, keeping code modular, and having good architecture. The team intentionally kept it a research preview to gather feedback on environment customization and pricing, with plans to integrate seamless transitions between cloud and CLI.

The #1 SWE-Bench Verified Agent

Apr 2, 2025 · 31:46

Guy Gur-Ari, co-founder of Augment Code, explains how their coding agent achieved the #1 spot on SWE-Bench Verified without fine-tuning, using off-the-shelf models with techniques like sequential thinking, ensembling, and code base understanding. He shares that orientation and memories help the agent learn codebase conventions, and that MCP support allows extensibility. Gur-Ari contrasts Augment's IDE-first approach with competitors like Devin and Factory, emphasizing meeting developers where they are. He also discusses the future of IDEs, the role of custom models, and recommends the SWE-RL paper for reinforcement learning in coding. The episode features a live demo of the agent implementing a feature from a Linear ticket and creating a PR.

Fullstack-Bench: The Eval for Coding Agents — with Sujay Jayakar, Chief Scientist, Convex

Mar 19, 2025 · 31:13

Sujay Jayakar, Convex chief scientist, presents Fullstack-Bench, a benchmark evaluating how well AI coding agents build full-stack apps on Convex, Supabase, and FastAPI+Redis, finding Convex’s tight feedback loops and strong abstractions let agents autonomously complete tasks within time limits. The benchmark tests three tasks (chat, to-do, files) across three backends with 30-60 minute limits and human hints. Convex succeeded autonomously on chat and to-do tasks, while Supabase agents struggled with RLS policy debugging and FastAPI agents with manual SSE wiring. Sujay discusses how platform design impacts AI code generation: simple procedural code, strong abstractions, and guardrails like end-to-end type safety improve LLM performance. He notes that API choices matter—things too similar to Firebase but not exact cause hallucinations—and that models like Claude 3.7 sometimes regress on platform-specific evals. The episode concludes with Convex’s focus on enabling AI-generated backends and anticipation of new agent workflow patterns.

Beating OpenAI and Anthropic by Looking At Data: the new #1 on SWE-Bench w/ W&B CTO Shawn Lewis

Jan 28, 2025 · 34:31

Shawn Lewis, CTO of Weights & Biases, built a coding agent using OpenAI's o1 model that achieved 64.6% on SWE-bench verified, the top score at the time. He developed a TypeScript agent framework called PhaseShift and an Eval Studio tool backed by Weave, emphasizing that rigorous data inspection and tooling are critical for agent development. Lewis explains his process of running parallel rollouts and a crosscheck method to select the best trajectory, which contributed a 6% improvement. He also discusses the challenges of adapting reasoning models like o1 for agentic tasks and shares plans to open-source PhaseShift. The episode highlights Lewis's dogfooding approach, using Weights & Biases' own tools to build and evaluate the agent.

Best of 2024 in Agents (from #1 on SWE-Bench Full, Prof. Graham Neubig of OpenHands/AllHands)

Dec 25, 2024 · 51:25

Professor Graham Neubig of CMU and AllHands AI, maintainer of the open-source coding agent framework OpenHands, delivers a talk on the state of agents in 2024, arguing that agents are now capable of automating significant portions of software development but still struggle with information gathering and error recovery. He demonstrates OpenHands solving real coding tasks live, explains key design choices like using Python code execution over granular tool calls to reduce LLM invocations, and shares that Claude currently outperforms other models for agentic tasks due to superior instruction following and error correction. Neubig discusses eight perennial agent problems including agent-computer interface, human-agent interface, model selection, planning, workflows, exploration, search, and evaluation. He predicts every major LLM trainer will focus on agent-oriented models by mid-2025, that agent benchmarks like SWE-Bench and WebArena will saturate and require harder successors, and that the biggest challenge will be designing human-agent interfaces for systems that succeed only 30-40% of the time autonomously but 80-90% with human feedback. The talk also covers workflow memory as a…

Windsurf: The Enterprise AI IDE

Dec 13, 2024 · 1:06:36

Varun Mohan and Anshul of Codeium explain why they built Windsurf, a new AI IDE, arguing that VS Code's API limitations prevented them from delivering the best agentic experience. They detail how Cascade, their agentic system, uses proprietary retrieval and planning models alongside third-party LLMs, and describe their evaluation method that masks commits and tests incomplete code states. The duo also reveals that over 800,000 developers use Codeium extensions, that they still support JetBrains and Eclipse for enterprise customers, and that they intentionally avoided a waitlist launch. They discuss the trade-offs of building first-party vs. third-party models, the importance of 'go slow to go fast' in enterprise infrastructure, and their belief that individual developer profits should come after building switching costs through superior product.

The new Claude 3.5 Sonnet, Computer Use, and Building SOTA Agents — with Erik Schluntz, Anthropic

Nov 28, 2024 · 1:11:08

Anthropic's Erik Schluntz explains how Claude 3.5 Sonnet achieved state-of-the-art 49% on SWE-Bench Verified by using a minimal agent framework that gives the model full control via tools like bash and file editing, letting it self-correct without hard-coded workflows. He details the importance of tool design over prompt engineering, why XML tags work well, and how computer use reduces integration friction by letting models interact with any browser-based interface. Schluntz also shares his robotics experience, noting that while LLMs and diffusion models are promising for general-purpose robots, reliability at 99.9% and hardware variability remain major hurdles, and he expresses skepticism about self-driving as a business due to high vehicle costs versus driver-equivalent revenue.

[Paper Club] SWE-Bench [OpenAI Verified/Multimodal] + MLE-Bench with Jesse Hu

Oct 19, 2024 · 1:01:40

Jesse Hu walks through SWE-Bench, SWE-Bench Verified, SWE-Bench Multimodal, and MLE-Bench — benchmarks that evaluate AI agents on real-world coding and ML tasks, from fixing GitHub issues to winning Kaggle competitions. SWE-Bench scrapes 2,000+ pull requests from 12 Python repos to get free unit tests, with leaderboard scores rising from ~2% to 43% thanks to improved agent scaffolding. SWE-Bench Verified uses human raters and three criteria—well-specified problem, valid test, difficulty on a 0–3 scale—to filter 500 solvable tasks. SWE-Bench Multimodal targets 617 UI tasks from 17 JavaScript/TypeScript libraries, each including an image, but achieves only 12% accuracy. MLE-Bench repurposes Kaggle competitions: agents given 24 hours and an A10 GPU, with o1-preview scoring bronze in 17% of competitions and surpassing grandmasters (7 gold medals vs 5 required), though each run costs $4,000.

Language Agents: From Reasoning to Acting — with Shunyu Yao of OpenAI, Harrison Chase of LangGraph

Sep 27, 2024 · 1:26:33

Shunyu Yao of OpenAI and Harrison Chase of LangChain discuss the evolution of language agents, arguing that combining reasoning and acting through techniques like ReAct remains foundational, while tool design and cognitive architectures like CoALA will shape the next wave. Shunyu traces his work from text games to ReAct, showing how thinking as an action improves reliability; Harrison explains LangChain's adoption and the shift to LangGraph for stateful orchestration. They emphasize agent-computer interfaces (ACI) as a neglected area, noting that SWE-Agent's success came from optimizing tools for models, not humans. The conversation covers memory types (semantic, episodic, procedural) and applies CoALA's three dimensions to current systems. They highlight promising applications: customer support, coding agents, and spreadsheet-style UX for batch operations. Shunyu stresses that better benchmarks like SWE-Bench (which solved 30% of GitHub issues) and TauBench (best model 48%) are critical, and that data—not architecture—will drive future model improvements.

State of the Art: Training 70B LLMs on 10,000 H100 clusters

Jun 25, 2024 · 1:32:05

Josh Albrecht, CTO of Imbue, and Jon Frankle, Chief AI Scientist of Databricks, detail the challenges of training 70B models on 10,000 H100 clusters, from stolen Infiniband cables to GPUs silently returning wrong math. Imbue open-sources infrastructure scripts, health checks, the CARBS hyperparameter optimizer, and cleaned versions of 11 benchmarks plus 450,000 human judgments on ambiguity. Databricks released DBRX, a 132B-parameter mixture-of-experts model with 36B active per token, and a text-to-image model trained exclusively on Shutterstock's curated dataset. They discuss typical 3% weekly hardware failure rates, the difficulty of MoE training pushing network bandwidth, and why eval datasets need human scrutiny—most benchmarks are saturated once ambiguous examples are removed. Both emphasize that success requires deep understanding from firmware to final loss, not just fancy libraries.