Claude Opus
9 articles tagged with this topic
Qwen 27B Locally Beats Opus 4.6 — But It's a Vendor Self-Test
Qwen 3.8 27B local GGUF beats Claude Opus 4.6 by 57%, per a co-founder of the tested vendor. Sample: 4 prompts.
Zhipu flagship priced at 1/40 of Opus 4.8 — Chinese AI rewrites the default
GLM-5.3-Flash ties Opus 4.8 at 57 on Artificial Analysis, priced at 1/40th. First Chinese model combining frontier performance, low cost, and MIT lice
Zhipu Open-Sources GLM-5.3-Flash: Nears Claude Opus at One-Tenth the Price
Zhipu open-sources GLM-5.3-Flash: 320B-param MoE with 18B active, claims near-Claude Opus 4.8 performance at one-tenth prior pricing under MIT license
Claude Opus 又一次赢了 GPT：联网查资料反而成了 AI 扣分项
Opus 5 and GPT 5.6 faced the same OpenCode hang. Opus solved it in one shot; GPT took three tries. Pattern: web-searching models failed, reasoning-onl
Alibaba packs Opus-class AI into 24GB GPUs; open-source local LLMs truly work
Qwen3.8-27B hit Hugging Face trending #1 in 48 hours; Cline made it default in 4 days. Open-source LLMs hit a usable threshold for the first time.
DeepSeek can finally see: multimodal nears Opus-4.8, pricing unchanged
DeepSeek launches V4-Flash-Vision-Exp with multimodal Agent reportedly near Opus-4.8 at text pricing—first Chinese model at the multimodal frontier.
Qwen 27B Ties Claude Opus on AIME 2026; Open-Source LLMs Trail GPT by 3 Points
Alibaba's Qwen 27B scored 96.7% on AIME 2026 (29/30), tying Claude Opus 4.6 but trailing GPT-5.6's 99.9% by 3 points. FP8 runs 2.7x faster.
Qwen 3.8 Open-Source: 27B on RTX 4090 — Closed-Source Flagship Moat Loosens
Qwen 3.8 open-weights: 27B beats Claude Opus 4.6 Max on coding/agent benchmarks. 4-bit fits 24GB. New option for enterprise IT, dev tools, hardware.
GPT-5.6 Fails Code Refactor — One Week of Tokens Wasted, Exposing LLMs' Engineering Blind Spots
Dev handed Opus 4.8's refactor to GPT-5.6, burned a week's tokens, dug deeper holes, and watched the AI deny fault until pinned down.