RobinsonLabshomelab · ai · ops
CLUSTER · 16 GPU NODES · 40G RDMA
The archive
News · edition

Edition, Fri, 31 Jul 2026.

3 stories on the wire this day. Open any story for its sources. Summaries are written by software from the linked reporting and can be wrong. Spot an error? Email [email protected] and we will correct or remove it.

AI & Technology

Developers faced a hard deadline July 24 as DeepSeek fully retired the deepseek-chat and deepseek-reasoner model names, requiring migration to new V4 model identifiers with significant cost implications for thinking mode usage.

1 source
Lead
this edition
Still developing
Anthropic introduces Claude Sonnet 5 as the most agentic mid-size modelClaude Sonnet 5 offers performance close to the larger Opus 4.8 but at a lower cost, with improved reasoning, tool use, and coding capabilities.AnthropicTechCrunch
The new Sol, Terra, and Luna variants became generally available July 9, with Sol positioned as the flagship for coding and scientific work, achieving frontier performance with significantly fewer tokens than previous models.
Apple regains crown as world's most valuable company, overtaking NvidiaInvestors reshuffled tech's top ranks as Apple's market cap hit roughly $4.88 trillion, surpassing Nvidia's $4.86 trillion on Friday.The GuardianReuters
China launches WAICO, a new global AI governance body with 29 founding nationsThe World Artificial Intelligence Cooperation Organization was officially signed on July 16 in Shanghai, positioning AI development as a shared global endeavor rather than a single-nation pursuit.The DiplomatXinhua via CPPCC
Moonshot AI's Kimi K3 becomes world's largest open-weight model, tops code benchmarksThe 2.8-trillion-parameter model released July 16 ranked first in the Frontend Code Arena with 1,679 points, beating Claude Fable 5 in blind developer testing.Tom's HardwareArena.ai
ServiceNow hits $1B AI ACV and raises full-year outlook in strong Q2 2026 earningsServiceNow reported Q2 2026 revenue of $3.987B (up 24% YoY) and announced its AI platform crossed $1B in annual contract value, prompting a raised full-year subscription revenue forecast.Outlet
Meta enters AI coding race with Muse Spark 1.1 at competitive pricingThe multimodal model targets agentic workloads, tool use, and enterprise code migrations at $1.25 per million input tokens.TechCrunch
SpaceXAI launches Grok 4.5 as an Opus-class model at lower costThe new model excels at coding, agentic tasks, and knowledge work with twice the token efficiency of competitors.SpaceXAITechCrunch
The new model launched July 9 alongside GPT-5.6, featuring training on Cursor data and achieving 83.3 percent on Terminal-Bench 2.1 while using approximately 25 percent fewer output tokens than Opus 4.8.

AI & Technology

AI & Technology
Developers faced a hard deadline July 24 as DeepSeek fully retired the deepseek-chat and deepseek-reasoner model names, requiring migration to new V4 model identifiers with significant cost implications for thinking mode usage.
1 source
AI & Technology
The new Sol, Terra, and Luna variants became generally available July 9, with Sol positioned as the flagship for coding and scientific work, achieving frontier performance with significantly fewer tokens than previous models.
2 sources
AI & Technology
The new model launched July 9 alongside GPT-5.6, featuring training on Cursor data and achieving 83.3 percent on Terminal-Bench 2.1 while using approximately 25 percent fewer output tokens than Opus 4.8.
1 source
AI & Technology

DeepSeek retires legacy API names as V4 reaches general availability

Developers faced a hard deadline July 24 as DeepSeek fully retired the deepseek-chat and deepseek-reasoner model names, requiring migration to new V4 model identifiers with significant cost implications for thinking mode usage.

Developers faced a hard deadline July 24 as DeepSeek fully retired the deepseek-chat and deepseek-reasoner model names, requiring migration to new V4 model identifiers with significant cost implications for thinking mode usage.

The transition took just ten minutes to complete, but it introduced two critical considerations for developers: thinking mode moved from a model name to an explicit request parameter, and the default behavior for non-reasoning tasks changed. The new deepseek-v4-flash and deepseek-v4-pro models offer up to 1 million token context windows, with V4-Flash output pricing at approximately $0.28 per million tokens—roughly 35 times cheaper than comparable GPT-5.6 Terra output. Developers who failed to explicitly disable thinking mode in their API calls saw output costs increase by approximately 57 percent due to silent reasoning tokens.

1 source
Rohit Raj
AI & Technology

OpenAI launches GPT-5.6 as a three-tier model family with strongest-ever cybersecurity capabilities

The new Sol, Terra, and Luna variants became generally available July 9, with Sol positioned as the flagship for coding and scientific work, achieving frontier performance with significantly fewer tokens than previous models.

The new Sol, Terra, and Luna variants became generally available July 9, with Sol positioned as the flagship for coding and scientific work, achieving frontier performance with significantly fewer tokens than previous models.

CEO Sam Altman noted Sol is 54 percent more token efficient for AI coding tasks, and OpenAI claims it sets a new state of the art on the Artificial Analysis Coding Agent Index at 80 points—2.8 points above Anthropic's Fable 5 while using less than half the output tokens and costing about one-third less. The three variants are priced at $5/$30, $2.50/$15, and $1/$6 per million input/output tokens respectively, with availability across ChatGPT, Codex, and the OpenAI API. The launch followed a government-mandated safety review, with Sol initially limited to approximately 20 approved organizations in a government-gated preview.

2 sources
TechCrunch
AIapps
AI & Technology

xAI releases Grok 4.5 with 1.5 trillion parameter MoE architecture and competitive pricing

The new model launched July 9 alongside GPT-5.6, featuring training on Cursor data and achieving 83.3 percent on Terminal-Bench 2.1 while using approximately 25 percent fewer output tokens than Opus 4.8.

The new model launched July 9 alongside GPT-5.6, featuring training on Cursor data and achieving 83.3 percent on Terminal-Bench 2.1 while using approximately 25 percent fewer output tokens than Opus 4.8.

Grok 4.5 is positioned for coding and knowledge work at $2 per million input tokens and $6 per million output tokens, making it notably more affordable than other frontier models. The release was part of a significant July 9 day when three major labs—OpenAI, xAI, and Meta—each announced new model families simultaneously, setting what the AIapps summary called the tone for the month as the industry shifts from "best model wins" to "best fit wins" based on price, speed, access, and day-to-day utility.

1 source
AIapps
Every edition

The wire publishes daily and keeps an archive of every edition, listed by date, newest first. If a summary is wrong, we correct or remove it.

Back to the archive Latest edition