Agents Last Exam
PulseAugur coverage of Agents Last Exam — every cluster mentioning Agents Last Exam across labs, papers, and developer communities, ranked by signal.
- instance of DeepSWE v1.1 90%
- instance of GPT 5.6 "Sol" 90%
- instance of GPT -6 Astra 90%
- competes with Claude Fable-5 80%
- used by Terminal-Bench 4.0 70%
- developed by Claude Fable-5 70%
- competes with GPT 5.6 "Sol" 70%
- developed Claude Fable-5 70%
- instance of Claude Opus-5 60%
- used by Claude Fable-5 50%
- 2026-06-11 research_milestone GPT-5.5 outperformed Claude Fable 5 on the new Agents Last Exam benchmark. source
8 day(s) with sentiment data
-
OpenAI launches GPT-6.1 Terra with 50% price cut and 3x speed boost
OpenAI has released GPT-6.1 Terra, a new default tier model that offers significant cost and speed improvements over previous versions. Priced at $1 per million input tokens and $5 per million output tokens, Terra is ov…
-
New 'Thinking-in-Bets' method boosts LLM agent reliability
A new method called "Thinking-in-Bets" has been developed to improve the reliability of language model agents. This training-free technique involves agents creating a private record of their claims and confidence levels…
-
MiniMax M3.1 Flash open-source model released with 428B parameters
MiniMax M3.1 Flash, a new open-source model, has been released with 428 billion parameters, activating 23 billion per token. The model demonstrates significant advancements in agentic tasks, coding, and cybersecurity, a…
-
DeepSeek V4.1-Flash-1017 offers benchmark gains with shorter outputs at unchanged API prices
DeepSeek has released V4.1-Flash-1017, an updated version of its V4.1-Flash model, available under the MIT license. This new iteration shows performance improvements across 11 benchmarks, with gains ranging from 2.0 to …
-
MiniMax M3.1 Flash launches with lower prices, API-only access
MiniMax has launched its M3.1 Flash model, offering a significant price reduction compared to its predecessor, M3, with input tokens priced at $0.15 and output tokens at $0.60 per million. While MiniMax claims a 70% cos…
-
DeepSeek V4.1 Flash 1017 debuts with major gains in coding and agent tasks
DeepSeek has released an updated version of its open-source large language model, DeepSeek V4.1 Flash 1017, on October 17, 2026. This iteration features a significant improvement across various benchmarks, particularly …
-
MiniMax launches 428B M3.1 Flash MoE model with 1M context
MiniMax has launched its M3.1 Flash model, a 428B parameter Mixture-of-Experts model with a 1 million token context window. The model is available via the MiniMax API and is designed for agentic tasks, coding, and visua…
-
DeepSeek releases V4.1-Flash-1017, boosting agent performance and cutting costs · 9 sources tracked
DeepSeek has released V4.1-Flash-1017, an updated open-weight model that offers improved performance and cost efficiency for agentic tasks. This new checkpoint boasts higher scores across 11 reported benchmarks, faster …
-
AI benchmark scores lack clarity; new checklist proposed
A new checklist for evaluating AI benchmark scores has been proposed, highlighting inconsistencies and potential manipulation in how AI models' computer use capabilities are measured. The author points out that leading …
-
OpenAI unveils GPT-6 Astra, claiming new state-of-the-art across benchmarks · 6 sources tracked
OpenAI has announced GPT-6 Astra, its latest state-of-the-art model. The company claims Astra excels across a wide range of benchmarks, including FrontierMath Tier 4, ARC-AGI 3, and TerminalBench-4.0, demonstrating sign…
-
DeepSeek V4 multimodal model weights released for inspection
DeepSeek has released the weights and reference code for its V4 multimodal model, allowing researchers to examine its visual processing capabilities. Unlike simple image-to-text additions, V4 integrates visual tokens di…
-
Qwen3.8-27B open-source model tops leaderboards with advanced agency
Alibaba's Qwen team has open-sourced Qwen3.8-27B, a 27-billion parameter model that has achieved top rankings on several benchmarks, including SWE Bench Pro and OSWorld. This model demonstrates significant advancements …
-
OpenAI's GPT-5.6 Sol excels at code generation but struggles with database population
OpenAI has released GPT-5.6 Sol, which demonstrates significant improvements in coding tasks and token efficiency, outperforming previous models like Claude Opus 4.8 in benchmark tests. However, the model struggles with…
-
AI Agents Last Exam Leaderboard Nearing Saturation by February
The Agents Last Exam leaderboard is nearing saturation, with current benchmarks indicating it will be fully saturated by February of next year. This leaderboard tracks the performance of AI agents on various tasks, meas…
-
GPT 5.6-Sol outperforms Claude Fable-5 on agentic tasks, but questions remain
A new model, GPT 5.6-Sol, has reportedly surpassed Anthropic's Claude Fable-5 on long-horizon agentic tasks, achieving a score of 53.6 compared to Fable-5's 40.5 on the Agents' Last Exam. Despite this performance edge a…
-
OpenAI launches GPT-5.6 family with Sol, Terra, and Luna tiers
OpenAI has launched its new GPT-5.6 model family, featuring three tiers: Sol (flagship), Terra (balanced), and Luna (fastest and cheapest). These models offer varying price-performance points, with Sol being the most ca…
-
Meta AI hires AI safety expert Jianfeng Gao to lead Superintelligence Labs
Meta AI has hired Dr. Yann LeCun's former colleague, Dr. Jianfeng Gao, to lead its Superintelligence Labs. Gao, previously a professor at UC Berkeley and co-founder of Virtue AI, will focus on AI safety and security eff…
-
New Benchmark Shows GPT 5.5 Outperforming Claude Fable 5 on Real-World Tasks
A new benchmark called Agents' Last Exam (ALE), developed by researchers from UC Berkeley and other institutions, has revealed surprising results in AI agent performance. In the most challenging tasks, leading models li…
-
GPT-5.5 Outperforms Claude Fable 5 on New AI Agent Benchmark
OpenAI's GPT-5.5 has reportedly outperformed Anthropic's Claude Fable 5 on the new Agents' Last Exam (ALE) benchmark. This benchmark, developed by UC Berkeley, evaluates AI agents' ability to perform complex, multi-step…
-
GPT-5.5 surpasses Claude Fable 5 on new AI agent benchmark
OpenAI's GPT-5.5 has outperformed Anthropic's Claude Fable 5 on a new AI benchmark called Agents Last Exam (ALE). This benchmark, developed by Berkeley RDI with input from over 300 experts, tests autonomous AI agents. T…