# AI News Digest - 2026-06-29

1. Coinbase switched to Chinese AI models such as GLM 5.2 and Kimi 2.7, deployed an automated routing system to select models by task and price, and increased caching hit rates from 5% to 60%, which enabled the company to cut its AI spending by half despite rising token usage.

2. Anthropic's Fable 5 was expected to return within days as the Trump administration prepared to lift the restrictions imposed on June 12, subject to sign-off from the Pentagon and NSA.

3. Princeton researchers running CEO-Bench found that only three AI models finished above starting capital in a 500-day simulated startup test, with most models going broke and a simple rule-based heuristic outperforming nearly all AI agents.

4. Sina's VibeThinker-3B matched substantially larger models on math and coding benchmarks despite having three billion parameters, with researchers attributing the results to multi-stage post-training and proposing that logical reasoning compresses better than broad factual knowledge.

5. 360 founder Zhou Hongyi presented two AI security tools intended to compete with Anthropic's Mythos, reported that one tool had already flagged 3,432 vulnerabilities, and framed the strategic AI-security race as akin to a "cyber-nuclear" deterrent while noting Chinese models trailed Western ones by 20–30%.

# References

1. [https://the-decoder.com/coinbase-joins-the-rush-to-chinese-ai-models-as-western-labs-face-a-pricing-stress-test/](https://the-decoder.com/coinbase-joins-the-rush-to-chinese-ai-models-as-western-labs-face-a-pricing-stress-test/)

[Coinbase joins the rush to Chinese AI models as Western labs face a pricing stress test](https://the-decoder.com/coinbase-joins-the-rush-to-chinese-ai-models-as-western-labs-face-a-pricing-stress-test/)

1. [https://the-decoder.com/anthropics-fable-5-could-return-within-days-as-trump-administration-prepares-to-lift-restrictions/](https://the-decoder.com/anthropics-fable-5-could-return-within-days-as-trump-administration-prepares-to-lift-restrictions/)

[Anthropic's Fable 5 could return within days as Trump administration prepares to lift restrictions](https://the-decoder.com/anthropics-fable-5-could-return-within-days-as-trump-administration-prepares-to-lift-restrictions/)

1. [https://the-decoder.com/only-three-ai-models-finished-above-starting-capital-in-a-500-day-startup-survival-test/](https://the-decoder.com/only-three-ai-models-finished-above-starting-capital-in-a-500-day-startup-survival-test/)

[Only three AI models finished above starting capital in a 500-day startup survival test](https://the-decoder.com/only-three-ai-models-finished-above-starting-capital-in-a-500-day-startup-survival-test/)

1. [https://the-decoder.com/sinas-open-model-vibethinker-3b-aims-to-show-reasoning-compresses-well-but-factual-knowledge-doesnt/](https://the-decoder.com/sinas-open-model-vibethinker-3b-aims-to-show-reasoning-compresses-well-but-factual-knowledge-doesnt/)

[Sina's open model VibeThinker-3B aims to show reasoning compresses well but factual knowledge doesn't](https://the-decoder.com/sinas-open-model-vibethinker-3b-aims-to-show-reasoning-compresses-well-but-factual-knowledge-doesnt/)

1. [https://the-decoder.com/chinese-cybersecurity-firm-builds-ai-tools-to-rival-mythos-and-frames-the-race-as-cyber-nuclear-deterrence/](https://the-decoder.com/chinese-cybersecurity-firm-builds-ai-tools-to-rival-mythos-and-frames-the-race-as-cyber-nuclear-deterrence/)

[Chinese cybersecurity firm builds AI tools to rival Mythos and frames the race as cyber-nuclear deterrence](https://the-decoder.com/chinese-cybersecurity-firm-builds-ai-tools-to-rival-mythos-and-frames-the-race-as-cyber-nuclear-deterrence/)

For the site tree, see the [root Markdown](https://ixtj.dev/.md).