
MAI Code 2 Launch: Price vs Benchmark Reality for Enterprise Coding
Microsoft's MAI Code 2 hits 98% of GPT-6.1 Sol's score at 40% of the cost, but the promo price doubles in May 2027. INR rates and GST included.
for the people who have to decide about it
Clear, hype-free guides to using AI in your work and business. What actually works, tested and explained in plain English.

Microsoft's MAI Code 2 hits 98% of GPT-6.1 Sol's score at 40% of the cost, but the promo price doubles in May 2027. INR rates and GST included.

Google launched Gemini 4 Neon on Nov 18, 2026 at $3/$15 per million tokens with a 1M context. AA Index 61 ranks it #3, and it tops the coding-agent index.

A 1.8M-parameter energy-function reasoner from Duke, UIUC and Harvard hits 99.9% on Sudoku-Extreme with 345x less inference compute. MIT code, trains in hours.

Japan's KARTE breach: vendor PLAID lost 79,249 logins and integration keys; JINS, DHC and MUFG disclose 44 million records. Timeline, impact and what to revoke.

OpenAI says its models rewrote AMD GPU RTL for 14% PPA gains and silicon-proven Jalapeno savings, as GPT-Synopsys industrializes AI chip design.

China Telecom's Xingchen AGI Lab trained a 285B MoE on 8,192 Huawei Ascend 910C chips with zero Nvidia silicon. Its open 29B member runs on one GPU.

A new preprint shows Qwen3.5-27B diagnosing and editing its own weights with no gradients, gaining 6.6 points on held-out problems - and smaller models fail.
The Tech Archive covers artificial intelligence for people who have to make decisions about it. Every piece starts from something checkable: a model release and what changed in it, a tool and what it costs to run, a research result and what it does and does not show. We write about model releases, AI agents, developer tooling, research, funding, and policy, and we link to the primary source so you can disagree with our reading of it.
What you will not find is a launch rewritten from a press release, a benchmark quoted without its conditions, or a prediction dressed up as a finding. When something is genuinely unclear we say so and explain what evidence would settle it. That is slower than the news cycle and it is the point.
Practical automation for teams without an engineering department: what to adopt, what it actually costs once you run it every day, and which tasks are still cheaper to do by hand. Read the small business guides.
How generative tools change research, drafting, and measurement, and where they quietly produce work that reads fine and says nothing. Read the marketing guides.
Agent frameworks, model APIs, and local inference, tested on real work rather than demos, with the failure modes written down beside the setup instructions. Read the developer guides.
New articles publish several times a week. You can follow along by RSS, browse everything in the full article archive, or start from a topic.