DeepSeek's cheap model just beat Claude and GPT at coding
A 763B open-weight budget model tops four agentic benchmarks while costing $0.18 per task.
2026-10-187 views
A 763B open-weight budget model tops four agentic benchmarks while costing $0.18 per task.
Microsoft completed pre-training on MAI-Base-2, a 2.8T-parameter MoE model. Its AI chief claims it outperforms GPT-5.6's base model using only a quarter of the compute.
After two weeks behind the Fairwind Program, Google’s frontier model goes wide to developers, enterprises, and subscribers — with a 1M token limit and aggressive pricing.
Meta's first frontier-tier model beats Claude at long coding, but the version you can buy trails Opus 5.5 at the same price.
Anthropic's smallest model now outperforms OpenAI's GPT-6 Luna on nearly every benchmark at one-tenth the price.
Mistral AI launches Mistral Large 4, a 1.05T parameter MoE model with open weights coming October 27, positioning it as Europe's most powerful open-weight AI.
Google's new frontier model delivers 1M token output, SOTA coding at 77.9% DeepSWE, and half-price pricing vs Claude Opus 5.5.
Anthropic drops a mid-tier model that beats GPT-6 Sol on OpenAI's own benchmark, costs half of Opus, and lands the day before DevDay.