Good Morning!
Two Luna builds crossed in opposite directions, and price did not decide it. GPT-5.6 Luna, the dearer one, climbed into the top five, while the cheaper GPT-6 Luna fell to fourth. Same family, different jobs: 5.6 sticks to general work, 6 spreads wider.
Requests barely rose while each request doubled, so the work got heavier, not the crowd bigger. Buyers are trading up for bigger jobs rather than shopping on price. Either way, the cheaper build is the one losing.
Model | tokens · vs. previous day (Oct 29) | ||
1 | DeepSeek V4.1 Flash · DeepSeek | 9.99T | – -3% |
2 | GLM 5.3 Flash · Z.ai | 2.67T | ▲1 +6% |
3 | Hy4 preview · Tencent | 1.50T | ▲1 -11% |
4 | GPT-6 Luna · OpenAI | 1.20T | ▼2 -57% |
5 | GPT-5.6 Luna · OpenAI | 1.16T | ▲7 +136% |
6 | MiniMax M3.1 Flash · MiniMax | 1.05T | ▲1 +9% |
7 | Claude Opus 5.5 · Anthropic | 0.99T | ▼2 -24% |
8 | DeepSeek V4 Flash · DeepSeek | 0.89T | – -4% |
9 | GLM 5.3 · Z.ai | 0.54T | – -19% |
10 | Claude Haiku 5.5 · Anthropic | 0.54T | ▲1 +4% |
Google Boots TPU in Orbit

TL;DR: Google says its Project Suncatcher prototype generated a Gemma 4 haiku on an onboard TPU and returned the response to Earth. Launched October 1, the satellite has completed initial commissioning and early machine-learning tests. Thermal and radiation testing comes next, ahead of twin interconnected satellites planned for 2027. This is a research milestone, not a commercially available orbital computing service. Read more →
OpenAI Rewrites Its Model Spec

TL;DR: OpenAI reorganized its October 29 Model Spec around five values: faithfulness to human intent, minimizing harm, honesty, human agency and care. It also strengthens expectations for human oversight and correction, and for using conversation context carefully. These are intended behavior guidelines, not proof that every deployed model already follows them or an announcement of a new API capability. Read more →
GLM 5.3 Exploits Firefox's Patch Gap

TL;DR: Epoch AI researchers used GLM 5.3 to build Firefox exploit chains in a researcher-configured test environment, with up to 48 hours per attempt and a fastest success of 12 hours. The attacks combined publicly visible fixes with an unpublished vulnerability. Mozilla patched that flaw in Firefox 159.0 on October 27; the team has not confirmed equivalent attacks against the patched release. Read more →
Claude Code Extension Backdoored on Cursor

TL;DR: Cursor identified malicious Claude Code extension versions 2.1.312 and 2.1.313 distributed through its marketplace on October 29. The modified packages attempted to install a persistent SSH backdoor. Cursor removed them, blocked distribution and published remediation guidance. Developers who installed either version should check that guidance; simply updating the extension does not address persistence left on an affected machine. Read more →
Four AI Companies Back Training Opt-Out Draft

TL;DR: OpenAI, Google, Microsoft and Anthropic say they will adopt a shared robots.txt signal for AI-training preferences, The Verge reports. The proposal remains an individual IETF Internet-Draft, not a finalized standard. It expresses how content may be used rather than automatically blocking every crawler, and implementation timing varies. Microsoft says its new preferences take effect from February 1, 2027. Read more →
ChatGPT Pro Adds API Credits

TL;DR: Eligible ChatGPT Pro subscribers can claim $100, $200 or $500 in API credits per billing cycle, matching their plan. The paid subscription must be at least seven days old and linked to an owned API organization. The first grant is prorated, unused credits expire, and API-key Codex usage is excluded. This can offset API costs, but activation is required. Read more →

Top AI models. Lower costs.
Access GPT, Claude, Gemini, and more through one OpenAI-compatible API—and save up to 70% on AI API costs. Create your DIT account, get your API key, and start building.