dropkit.contents

Ledger on GitHub. Hub here. Branches elsewhere.

← Drops · 2026-09-11

drop-2026-09-11-kr-pm

claimlabelordernumbersURL
GPT-5.6 Sol led Terminal-Bench 2.1 in an August 2026 benchmark roundup.confirmed🟢 robust89.5https://www.morphllm.com/best-ai-coding-agents-2026
Claude Opus 5 led SWE-bench Verified in August 2026 leaderboards.confirmed🟢 robust97.00%https://leaderboard.steel.dev/leaderboards/swe-bench-verified/
Grok 4.6 was the best newcomer in Agents on Rails, behind GPT-5.6 Sol.confirmed🟢 robust52/63 runs; 33% recall; $49 campaign costhttps://rubyonrails.org/2026/8/17/agents-on-rails-grok-4-6-glm-5-3-gemini-3-7-flash-and-opus-4-8
Grok 4.6 pricing stayed at $2 input and $6 output per 1M tokens.confirmed🟢 robust$2 / $6 per 1Mhttps://rubyonrails.org/2026/8/17/agents-on-rails-grok-4-6-glm-5-3-gemini-3-7-flash-and-opus-4-8
Perplexity API added Agent API, Search API, Embeddings API, and Sandbox API.confirmed🟢 robust4 APIshttps://releasebot.io/updates/perplexity-ai
Claude Opus 5 led SWE-bench Verified in secondary August 2026 summaries.partial⚠️ sensitive96.0%–97.0%https://localaimaster.com/models/swe-bench-explained-ai-benchmarks
Claude Fable 5 was listed as a coding contender in August 2026 comparison pages.partial⚠️ sensitive95.0% SWE-bench Verified; 80.0% SWE-bench Prohttps://benchlm.ai/blog/posts/claude-opus-5-benchmarks
Gemini 3.1 Pro was paired with a 1M-token context and mid-tier pricing in comparison pages.partial⚠️ sensitive$2.00 / $12.00https://tech-insider.org/ca/claude-opus-5-vs-gpt-5-6-sol-vs-qwen-3-8-max-2026/
Grok 4.6 latency was reported as slower than faster models in benchmark writeups.partial⚠️ sensitive14.51s TTFB; 43.82s TTFB; 32.30s TTFBhttps://codersera.com/blog/grok-4-6-launch-guide-2026/
GLM-5.3-Flash was reported as a fast frontier model in August 2026 coverage.partial⚠️ sensitive84.3 Terminal Benchhttps://aitoolsrecap.com/Blog/ai-news-august-28-2026
Tencent Hy4 preview was reported with strong benchmark scores and lower pricing.partial⚠️ sensitive92.3 GPQA Diamond; 65.7 SWE-bench Pro; $0.834 / $2.501 per 1Mhttps://www.buildfastwithai.com/blogs/tencent-opens-a-770b-model-under-apache-2-0-ai-news-aug-31
Qwen3.8-Max was reported near the top of open-weight frontier comparisons.partial⚠️ sensitive86.6 Terminal Bench 2.1; 92.6 GPQA Diamondhttps://aitoolsreview.co.uk/insights/qwen-3-8-max
DeepSeek release notes listed Terminal-bench performance in low-30s.partial⚠️ sensitive31.3https://releasebot.io/updates/deepseek