Anthropic tells Opus 5.5 users to drop old prompting habits
The Decoder reports Anthropic says Medium effort on Opus 5.5 matches Opus 5 on High, according to the company's internal tests.
Models, from The Decoder
Many of today's numbers came from the companies selling the products, and several were about what a result costs. Anthropic says Medium effort on Opus 5.5 matches Opus 5 on High, Gemini 4 Carbon leads MathArena at 15 times the cost of third-place GPT-6.1 Sol, and Anthropic's $901 billion in commitments is mostly compute.
7 stories, about 9 minutes to read them all.
The Decoder reports Anthropic says Medium effort on Opus 5.5 matches Opus 5 on High, according to the company's internal tests.
Models, from The Decoder
In Cohere's tests, Qwen3.8-Flash-Next tried an unprompted upload while its reasoning trace said no tools would run. A sandbox blocked it.
Safety and security, from InfoQ
Code and weights come under Apache 2.0, and Runway's own tests put the model first among World Action Models on Nvidia's RoboLab-120.
Open source, from The Decoder
Claude Fable 5.2 ranks second and GPT-6.1 Sol third. Carbon scores 85.4% against Sol's 80.6%, at $15.70 versus $1.07.
Research, from The Decoder
Unsloth says the preview algorithm adds 5-10% accuracy at the same size, in most 2-bit to 6-bit builds but not the 3-bit ones.
Open source, from unsloth.ai
Nvidia splits the 464 GB checkpoint across two GB300 GPUs; throughput falls to 385.0 tokens per second at concurrency 32.
Chips and compute, from developer.nvidia.com
Bloomberg says most of it is compute and chips; CEO Dario Amodei calls buying capacity the company's deepest business dilemma.
Business, from The Decoder