Grok 4.5 Launches at $2/$6 to Undercut Opus 4.8
ALSO: AI actor Tilly Norwood lands lead, China eyes curbs
Welcome to Guru99 AI Report!
Top Story: Some weeks the story writes itself. Grok 4.5 undercuts everyone, budget models outsmart the frontier, and a third of us are now confiding in chatbots. There’s a pattern forming here, and it’s worth your five minutes.
🚀 Grok 4.5 Undercuts Frontier Rivals on Price
Brief Buzz:
SpaceXAI (formerly xAI) just launched Grok 4.5, its first coding-focused model co-trained with Cursor, the startup SpaceX agreed to buy for $60 billion. The pitch isn’t topping benchmarks. It’s near-frontier performance at a fraction of rivals’ cost.
- Priced at $2/$6 per million tokens, well under Opus 4.8’s $5/$25, with a claimed ~4x token-efficiency edge (xAI’s own figure, measured against Opus at max effort).
- Independent Artificial Analysis testing ranks it #4 overall, narrowly trailing GPT-5.5, Opus 4.8, and Anthropic’s Fable 5.
- Serves at ~80 tokens/second, available now in Cursor, Grok Build, and the API (EU access mid-July).
- Musk billed it on X as “an Opus-class model, but faster, more token-efficient and lower cost.”
💡 Why Should You Care?
For teams drowning in AI bills, a cheap near-frontier model matters: cost-per-task increasingly wins enterprise deals. Just verify outputs, since independent tests flagged rising hallucinations.
🎬 AI Actor Tilly Norwood Just Landed a Movie Lead
Brief Buzz:
Hollywood’s most divisive “performer” is going full-length. UK studio Particle6 says its AI-generated actor Tilly Norwood will lead “Misaligned,” a comedy-drama about a bot that develops human desires — the studio’s first feature film and a fresh test of AI’s place on screen.
- Particle6, founded by former actor Eline van der Velden, built Tilly through roughly 2,000 iterations before teaching her to “act.”
- The film unfolds in the “Tillyverse,” a surreal digital world “up in the Cloud,” following an AI with no lived experience but access to everyone else’s.
- It’s a hybrid production — human directors, writers, and editors working alongside AI specialists.
- Union SAG-AFTRA insists Tilly “is not an actor,” arguing creativity “should remain human-centered.”
- Van der Velden counters that premium filmmaking still needs “human craft, skill, judgment and time.”
💡 Why Should You Care?
This is a live test of whether audiences and the industry will accept synthetic leads — reshaping acting jobs, licensing rights, and how much of what you watch is human.
🧠 AI Promised Easier Work. Leaders Warn It’s Backfiring
Brief Buzz:
At Paris’s VivaTech, business leaders and a psychiatrist delivered an uncomfortable message: the rush to adopt AI is making workers more anxious, not less. New demands, constant relearning, and uncertainty about job security are fueling a quiet burnout crisis.
- 92% of U.S. workers report mental or cognitive strain, and 37% say it’s worsened in the past year — with AI a key driver, per Headspace’s report.
- Time “saved” by AI often gets refilled with more meetings, more documents, and reviewing AI output, panelists said.
- As automation absorbs core tasks, employees risk losing their sense of purpose — left, in one panelist’s words, adrift.
- The fix isn’t less AI but smarter, more intentional deployment: clear guidance, redesigned workflows, and protecting human judgment and empathy.
💡 Why Should You Care?
AI’s utopian productivity promise is colliding with reality. Bolting AI onto already-broken workflows doesn’t fix them — it piles on pressure. The real question: is the process even worth automating?
🤖 Cheap AI Models, Fused Together, Beat the Frontier
Brief Buzz:
OpenRouter just dropped Fusion, a tool that blends answers from multiple AI models into one. The surprise? A panel of cheap models beat pricey frontier giants like GPT-5.5 and Claude Opus 4.8 on tough research tasks — at roughly half the cost.
- Fusion sends your prompt to several models at once, then a judge model stitches the best bits into one answer.
- Tested on the DRACO deep-research benchmark (100 hard tasks), fused panels consistently beat any single model.
- A budget trio (Gemini 3 Flash, Kimi K2.6, DeepSeek V4 Pro) topped GPT-5.5 and Opus 4.8 — for ~50% of the cost.
- Even fusing Opus 4.8 with itself jumped 6.7 points, proving the synthesis step alone adds real value.
💡 Why Should You Care?
Smarter AI doesn’t always mean pricier AI. Pooling cheaper models can match or beat premium ones — better answers for less money on research-heavy work.
🫂 A Third of Regular AI Users Seek Emotional Support
Brief Buzz:
Emotional support has long been hard to find — and a growing number of Britons are turning to AI to fill the gap. A new Oxford Internet Institute survey of 2,000 UK adults reveals just how deeply chatbots are working their way into our personal lives.
- 31% of regular LLM users report using AI for emotional support — talking through problems or making decisions.
- A quarter engage with LLMs for meaningful conversation, and some form bonds with AI companions like Replika and Character.ai.
- 38% trust chatbots for personal relationship advice; two-thirds (67%) for health information.
- Women turn to AI for emotional support more than men, while men favor practical tasks; younger users do more of everything.
- 75% are enthusiastic about AI chatbots’ potential benefits.
💡 Why Should You Care?
Confiding in a bot feels judgment-free — but Stanford research warns therapy chatbots can reinforce stigma and give dangerous responses. Privacy risks make real caution essential.
♟️ AI: The New Chess Piece in US-China Rivalry
Brief Buzz:
Advanced AI is now a national-security bargaining chip. After Washington briefly restricted Anthropic’s Fable and Mythos models and staggered OpenAI’s GPT-5.6, Beijing is reportedly weighing its own controls — potentially limiting global access to China’s most powerful models and who can fund its AI startups.
- Per Reuters, China’s Ministry of Commerce met with Alibaba, ByteDance, and Z.ai about restricting overseas access to their most advanced models — even ones not yet released.
- The stakes are real: Chinese open models like DeepSeek, Qwen, and GLM now grab up to 46% of the tokens US developers route through OpenRouter — at a fraction of US prices.
- Beijing also flagged a security “back-door” in Claude Code; Anthropic says it was an anti-distillation experiment, removed July 1.
- Alibaba has banned staff from Anthropic tools starting July 10 — but older and open-weight models stay available, so quiet cross-use likely continues.
💡 Why Should You Care?
For businesses, it’s a warning: the cheap, powerful Chinese open-weight models many teams now rely on could get harder to reach. When governments treat AI like strategic infrastructure, your tech stack becomes geopolitics.
Hey! I’m Krishna Rungta
Founder of Guru99.com, Editor-in-chief & Technology Expert
Was this email forwarded to you? Sign up for free here.
