Search

AI News · 3 min read

Cognition Launches SWE-2 Flash: 276B-Parameter Coding Model at One-Fifth the Price of Flagship

Cognition's SWE-2 Flash runs at three times the speed of its flagship at one-fifth the cost, using a mixture-of-experts architecture with 12B active parameters.

Nov 5

AI News · 3 min read

Nanbeige Releases 33B Coder Under Apache 2.0 With Half-Million-Token Context

Chinese lab Nanbeige released a 33B coding model under Apache 2.0 with a 524K-token context window, beating the current open-weight leader Qwen3.8-27B on 18 of 20 benchmarks including a 54% lead on DeepSWE.

Oct 26

Science & Research · 2 min read

Moonshot AI Open-Sources Kimi K2.7-Code — A 1-Trillion-Parameter Model Up to 12x Cheaper Than GPT-5.5

Moonshot AI released Kimi K2.7-Code as a fully open-source 1-trillion-parameter coding model at $0.95 per million input tokens — up to 12x cheaper than GPT-5.5 and Claude Fable 5.

Jun 14

Work & Labor · 3 min read

Microsoft Replaces GPT-4 in GitHub Copilot With Its Own Model Starting August 2026

Microsoft is replacing GPT-4 Turbo in GitHub Copilot with Project Polaris, its own in-house mixture-of-experts coding model, starting August 2026 — the company's biggest step yet toward AI independence from OpenAI.

Jun 3

AI News · 3 min read

Four Chinese Labs Release Open-Weight Coding Models at One-Third the Cost of Western Frontier AI

Four Chinese AI labs — Z.ai, MiniMax, Moonshot, and DeepSeek — released open-weight coding models matching Western frontier performance at one-third the inference cost in May 2026.

May 9