Friday, 17 April 2026
This dated roundup collects the most interesting AI and technology developments found for Friday, 17 April 2026.
Policy & Ethics
Payward to acquire Bitnomial, creating a fully CFTC-licensed derivatives platform
Combination of Bitnomial’s decade-built US regulatory infrastructure and Payward’s global distribution, scale, and multi-brand operating model to create one of the most comprehensively regulated and vertically integrated U.S. derivatives platforms The post Payward to acquire Bitnomial, creating a fully CFTC-licensed derivatives platform appeared first on Kraken Blog .
Read moreKraken Parent Payward Agrees to Acquire Derivatives Exchange Bitnomial for $550 Million
The acquisition will give Kraken access to complete CFTC-licensed crypto derivatives infrastructure in the U.S.
Read moreElizabeth Warren Accuses SEC Chair Paul Atkins of Potentially Lying to Congress
Senator Elizabeth Warren said Paul Atkins may have intentionally misled Congress by pushing back over the SEC's dwindling enforcement actions.
Read moreResearch & Products
Claude Opus 4.7 won 69 of 100 blind evals against Opus 4.6, judged by GPT-5.4, Gemini 3.1 Pro, and DeepSeek V3.2
I ran 100 blind questions across 5 categories (code, reasoning, analysis, communication, meta-alignment) and had three independent judges from three different model families evaluate both responses. Each judge saw responses labeled A and B with randomized order. Majority vote decides the winner. Per-judge results: |Judge|Opus 4.7 wins|Opus 4.6 wins|Ties|4.7 win %| |:-|:-|:-|:-|:-| |GPT-5.4|69|30|1|69.7%| |Gemini 3.1 Pro|76|22|0|77.6%| |DeepSeek V3.2|38|54|5|41.3%| |Aggregate|69|**
Read moreIntel Arc Pro B70 Open-Source Linux Performance Against NVIDIA RTX & AMD Radeon AI PRO Review
The R9700 is about 30% more than the B70, but it's more than 30% better. Overall, I rather have a R9700 than a B70.
Read moreQwen 3.6 35B crushes Gemma 4 26B on my tests
I have a personal eval harness: A repo with around 30k lines of code that has 37 intentional issues for LLMs to debug and address through an agentic setup (I use OpenCode) A subset of the harness also has the LLM extract key information from reasonably large PDFs (40-60 pages), summarize and evaluate its findings. Long story short: The harness tests the following LLM attributes: - Agentic capabilities - Coding - Image-to-text synthesis - Instruction following - Reasoning Both models at UD-Q4_
Read moreAnthropic’s Alarming Mythos Findings Replicated With Off-the-Shelf AI, Researchers Say
Security researchers used GPT-5.4 and Claude Opus 4.6 in an open-source harness to reproduce Anthropic's Mythos vulnerability findings for under $30 per scan.
Read moreDon't ask Qwen 3.6 35b to give you aski image of Yoshi :)
https://preview.redd.it/dfqed57qgsvg1.png?width=1706&format=png&auto=webp&s=3859209698d2e844e2731326e355d60928658f8a The most fun part was reasoning, here is a gist: https://gist.github.com/anzax/5f06716c66180013cd715f6c2e5848df There is a lot of criticism about Qwen 3.6 long reasoning, but actually I found it overthink for silly request like this, and in practical agentic tasks, my experience, it stays focused and
Read more