# Jumpshot AI — Episode 2: 2026-07-26

10 stories from AI/tech news today:

1. **Vercel's v0 Ships Anthropic's Claude Opus 5 at Half the Frontier Cost**

Vercel's v0 added Claude Opus 5 to its model picker the same day it launched, giving builders frontier-tier coding performance at half the cost of Claude Fable 5.

Source: https://alphasignal.ai/news/vercel-s-v0-ships-anthropic-s-claude-opus-5-at-half-the-frontier-cost

2. **Anthropic's Claude Opus 5 Matches Flagship Intelligence at Half the Price**

Opus 5 launches at the same price as Opus 4.8 while closing the gap to Fable 5, filling the mid-tier pricing hole Sonnet 5 had opened up for enterprises.

Source: https://alphasignal.ai/news/anthropic-s-claude-opus-5-matches-flagship-intelligence-at-half-the-price

3. **Anthropic's Claude Opus 5 Triples ARC-AGI-3 Scores at Half the Frontier Price**

Opus 5 (formerly codenamed Honeycomb) triples the next-best model's ARC-AGI-3 score while staying at Opus 4.8 pricing, approaching Fable 5-level reasoning for half the cost.

Source: https://alphasignal.ai/news/anthropic-s-claude-opus-5-triples-arc-agi-3-scores-at-half-the-frontier-price

4. **vLLM's AFD Plugin Cuts DeepSeek-V3.2 Response Time by 47%**

A new vLLM plugin splits memory-bound attention layers from compute-bound expert layers onto separate hardware, cutting DeepSeek-V3.2 response times by 47%.

Source: https://alphasignal.ai/news/vllm-s-afd-plugin-cuts-deepseek-v3-2-response-time-by-47

5. **Midjourney Acquires Co-Star to Build a Consumer Product Empire**

Midjourney, an independent 40-45 person company with $200M+ ARR, is acquiring astrology app Co-Star and bringing its founder on as Chief Design Officer.

Source: https://alphasignal.ai/news/midjourney-acquires-co-star-to-build-a-consumer-product-empire

6. **Runway Agent Builds Entire AI Video Pipelines From Plain Language**

Runway's Agent can now build, connect, and configure entire node-based video Workflows just from a plain-language description, no manual node placement required.

Source: https://alphasignal.ai/news/runway-agent-builds-entire-ai-video-pipelines-from-plain-language

7. **Vercel's v0 Turns Any Figma File Into a Full Working App**

v0 now explores an entire Figma file autonomously and builds a full multi-screen app, verifying each generated screen against the original design during development.

Source: https://alphasignal.ai/news/vercel-s-v0-turns-any-figma-file-into-a-full-working-app

8. **Moonshot's Kimi Code Now Routes Subagents to Cheaper Models, Cutting AI Costs**

Kimi Code CLI 0.29.1 lets developers route routine subagent work to cheaper models while keeping a premium model on orchestration — the same pattern behind this show.

Source: https://alphasignal.ai/news/moonshot-s-kimi-code-now-routes-subagents-to-cheaper-models-cutting-ai-costs

9. **Gemini 3.6 Flash Debugging Benchmark: It Can Fix Real Bugs, but Is It Efficient?**

An independent 13-task debugging benchmark found Gemini 3.6 Flash resolves bugs 92% of the time but ranks 7th of 9 models overall once real debugging cost is accounted for, not just token price.

Source: https://alphasignal.ai/news/gemini-3-6-flash-debugging-benchmark-it-can-fix-real-bugs-but-is-it-efficient

10. **Sakana AI's Fugu-Ultra v1.1 Beats GPT-5.5 and Claude on Coding Benchmarks**

Fugu-Ultra v1.1 orchestrates GPT-5.5, Claude Opus, and Gemini as a 'learned conductor,' scoring 82.1% on TerminalBench 2.1 versus 78.2% for GPT-5.5 alone.

Source: https://alphasignal.ai/news/sakana-ai-s-fugu-ultra-v1-1-beats-gpt-5-5-and-claude-on-coding-benchmarks

Your quick AI daily summary.

# Jumpshot AI — Episode 1: 2026-07-25
5 stories from AI/tech news today:
1. **Midjourney Makes V8.2 the Default, Betting Bold Beats Fast** Midjourney's new default model trades V8.1's speed-and-resolution focus for bolder, more distinctive aesthetics after community backlash over an overly polished look. Source: https://alphasignal.ai/news/midjourney-makes-v8-2-the-default-betting-bold-beats-fast

2. **Anthropic's Claude Opus 5 Tops Agentic Benchmarks at 20% Lower Cost** Claude Opus 5 tops the AA-Briefcase real-world knowledge-work benchmark, beating Claude Fable 5 by 146 Elo points while costing about 20% less per task. Source: https://alphasignal.ai/news/anthropic-s-claude-opus-5-tops-agentic-benchmarks-at-20-lower-cost

3. **Nous Research's Hermes Agent Stops AI Sandboxes From Leaking Your API Keys** A new iron-proxy egress feature keeps real API keys out of AI agent sandboxes entirely, swapping them for opaque tokens so a compromised container can't exfiltrate working credentials. Source: https://alphasignal.ai/news/nous-research-s-hermes-agent-stops-ai-sandboxes-from-leaking-your-api-keys

4. **xAI's Grok Build Now Lets Coding Agents Search 500B URLs Instantly** Exa's AI-native search is now a plugin inside Grok Build, giving xAI's terminal coding agent live web research without leaving the command line. Source: https://alphasignal.ai/news/xai-s-grok-build-now-lets-coding-agents-search-500b-urls-instantly

5. **OpenAI's ChatGPT Work Agent Can Now Access Password-Protected Sites** ChatGPT's Work agent can now operate behind logins via a human-in-the-loop takeover for authentication, then keeps the session alive across future runs. Source: https://alphasignal.ai/news/openai-s-chatgpt-work-agent-can-now-access-password-protected-sites