A post-midterm rout erased ~$2.5T from the Nasdaq 100 and put AI's money story on trial — OpenAI banked $11B at a discount, Anthropic's IPO walked into resistance, and agents kept shipping into the storm.
Announced Nov 4–5: a cyber-specialized variant of GPT-6 Astra built for authorized defensive work — vulnerability discovery, exploit development, web pentesting — with post-training to cut cyber refusals while keeping guardrails outside the scope.
Agents that own ongoing work: each gets its own computer inside Perplexity Computer, joins Slack channels, and hands tasks to other teammates. Bundled with the Max tier; research preview for individual subscribers.
Launch post: “Teammates work for you.”
Personal Finance in Claude links bank & card accounts via Plaid — read-only: “Claude can't move, manage, or invest your money.” Beta for US Pro/Max/Ultra. Plaid: 800+ AI companies now build on its rails.
Anthropic newsroom + Plaid blog, Nov 5
Amended IPO filing adds a striking risk factor: internal estimates “may imminently meet” its Responsible Scaling threshold for AI-driven acceleration — recursive self-improvement. No current model is designated as having crossed it.
Amended S-1, Nov 2 · DevX analysis
Moonshot AI's Kimi K3.1 Pro (launched Oct 23) landed #4 on the Artificial Analysis Intelligence Index — trailing only Claude Opus 5.5, Sonnet 5.5 and Muse Wonder Full. Once its weights drop (slated for this week; still unpublished as of Nov 6), it becomes the highest-scoring open-weights model anywhere.
Terminal-Bench 4.0 — vs 12.6% for Kimi K3.
BenchCAD with tools — vs Opus 5.5 at 96.2%, GPT-6 Astra at 95.9%.
per 1M tokens in/out · −33% vs Kimi K3
2.8T total params · 104B active · 1M context window
Google detailed the frontier's new price point: $2/$10 per 1M intro, then $4/$20; cached input 95% off. Cyber-defender-first access via the Fairwind Program; public GA not dated.
Reported to finish many tasks with ~half the tokens of Claude Sonnet 5.5.
blog.google Oct 28 · NeuralTrust / MindStudio Nov
Thinking Machines' new model scores 31 on the AA Intelligence Index — with weights promised, it would be the highest-scoring open-weights model developed outside China.
The open-weights wave now spans Moonshot, Baidu, Ricursive and Mistral's trillion-param Large 4 (Oct 30).
Artificial Analysis, Nov 6
| MODEL | IN | OUT |
| GPT-6.1 Terra | $1.00 | $5.00 |
| Gemini 4 Carbon (intro) | $2.00 | $10.00 |
| Kimi K3.1 Pro | $2.00 | $10.00 |
| GPT-6 Astra | $10.00 | $50.00 |
Agent labor keeps getting cheaper — Terra repriced computer-use at $1/$5 (Oct 28) and the floor hasn't held since.
Vendor price pages & prior issues; Carbon standard tier $4/$20
Wednesday Nov 5 was one of tech's worst single days on record: the Nasdaq Composite fell 5.52% to 24,315 — steepest since April 2025 — erasing ~$2.5T of Nasdaq-100 value and pushing the index into correction. Trigger: Democrats won both chambers on Nov 3, and investors repriced AI's regulatory and data-center risk overnight.
CFO Sarah Friar confirmed the price on camera and said the business grew 70% QoQ — but the discount is real, and the IPO has been delayed while the round completes.
Bloomberg via tastytrade/VT Markets · AP via ABC7 · 24/7 Wall St Nov 5–7
all-stock · equity ≈ $102.6B · close expected H1 2028
Day-one reads: MRVL −3% · LITE −6% — dot-com-flashback comparisons came free.