AI developer tools · What shipped

VIBE-Pro is a repo-level code generation and full-project delivery benchmark spanning web, mobile, and simulation-style implementation tasks, with data verified as of August 7, 2026. MiniMax M2.7 leads the public benchmark snapshot with a score of 55.6%, with 23 confirmed releases in the last 30 days. The benchmark tests whether models can complete substantial product requirements through end-to-end software delivery rather than single-file snippets, and BenchLM refreshes the benchmark quarterly.

Qwen3.7 Max leads the LiveCodeBench leaderboard as of August 7, 2026 with a 91.6% score, followed by Qwen3.7 Plus at 89.6% and GLM-4.7 at 84.9% across six tracked models. LiveCodeBench is a continuously updated coding benchmark built from newly collected LeetCode, AtCoder, and Codeforces problems, evaluating code generation, execution, test-output prediction, and self-repair on competitive programming tasks. The benchmark contributes 38% of the coding category score in BenchLM's overall model evaluation framework, with results verified and updated rolling through 2026.

LiveCodeBench v6, a named release benchmark for code generation, was updated August 7, 2026, with 16 AI models evaluated on competitive programming tasks. Sakana Fugu-Ultra leads at 93.2%, followed by Sakana Fugu at 92.9% and Kimi K2.6 at 89.6%. The benchmark measures contest-style code generation across a 21.2-point spread in the top-10 range and carries 20% weight in BenchLM.ai's overall scoring system, though it is currently displayed for reference only and excluded from the scoring formula.

Alibaba released Qwen 3.8-Max, a 2.4T parameter flagship model with 95 billion active parameters achieving 67% on SWE-Pro coding benchmark and 86% on OSWorld-Verified agentic tasks, with API pricing at $2/$6 per million tokens. Meta released Muse Spark 1.2, a coding-focused update featuring a 1M token context window and scoring 82.9% on Terminal-Bench 2.1 and 59.3% on Deep-SWE, priced at $1.25/$4.25 per million tokens; Meta also beta released Muse Code, a terminal-based coding agent for repository-scale engineering with multi-agent support. Minimax launched Minimax H3, an open-weights omni-modal model for audio-video generation achieving the number two spot on Arena for image-to-video, priced at $0.13 per second for 2K generation. Nvidia released Alpamayo 2 Super, an open 34B parameter reasoning vision-language-action model for autonomous vehicle development combining the Cosmos 3 Super Reasoner with a diffusion-based Action Expert. AWS announced general availability of Web Search on Amazon Bedrock for grounding foundation model responses in current web knowledge and released an automated web insight extraction solution using Amazon Bedrock AgentCore Browser. OpenAI solved ten major open mathematics problems using an internal version of its forthcoming Astra AI model, representing a transition to AI systems contributing genuinely new mathematics. Google expanded Ask Maps with agentic AI assistant tools for restaurant identification and food orders through Toast, Square, and Uber Eats, plus hotel price comparison and event ticket finding. AWS also released an event-driven architecture solution utilizing Amazon Bedrock and OpenSearch Serverless for AI-powered summaries from JavaScript-heavy web pages and RSS feeds.