- Sept 22: Opus 5.5 lands at $4/$20 per M tokens (20% below Opus 5) with cache reads down 60% to $0.20.
- Anthropic's benchmark table has it topping Fable 5.1 on agentic coding, computer use, and multidisciplinary reasoning; SWE-bench Pro 89.9%, Terminal-Bench 4.0 66.4%.
- Five-hour usage caps are gone for Pro, Max, Team, and seat-based Enterprise; rate-limit reset button ships alongside.
- Sonnet 5.5 and Haiku 5.5 are 'within weeks' per Anthropic.
- HN thread: 1,311 points, 861 comments — reaction splits between 'Fable at half the price' and reports of Max runs blowing their reasoning token budget.
- Sept 22: Sol is the balanced coding/agentic model at $2 in / $10 out per M tokens — half of GPT-5.6 Sol.
- 1.05M context, 128K max output, 922K single-input cap; DeepSWE 68.8%, OSWorld 2.0 64.4%, Artificial Analysis Intelligence Index 48.
- Luna is the cheap, low-latency sibling; both rolling out to Plus/Pro/Business/Enterprise/Edu in ChatGPT and to Copilot Pro+/Max/Business/Enterprise.
- Positioned below GPT-6 Astra on price — OpenAI's answer to Opus 5.5, dropped the same day.
- Bloomberg-obtained internal review: CENTCOM analysts trusted Maven Smart System's day-one target list built on 2018 imagery of a working elementary school.
- Two Tomahawks hit Shajarah Tayyebeh Elementary in Minab on Feb 28, killing 150+ including 123 children.
- CENTCOM's civilian-harm mitigation team had been cut from 10 to 1 under Hegseth's 90% staff reduction; no CHM review of Minab happened.
- Palantir's fix: revised target-refresh workflow, new civilian-movement OSINT feeds, dozens of Maven upgrades.
- Top HN thread of the day on the AI-in-warfare accountability question; picked up by Bloomberg, Gizmodo, Slashdot within hours.
- Browser Use's open-source web agent replaces every step-decision LLM call with one Jev typed-decision call — no generated text.
- ~17K stars in seven days; the token 'jev' now appears in 74 separate repo names in the same weekly feed.
- Positions Jev as a function-call primitive: unstructured state in, typed probabilistic decisions out, at 40–200x LLM speed.
- Sits alongside Fireship's Sept 21 Jev explainer as the moment TypeSafe's model crossed into mainstream developer awareness.
- Open-source Kev-4B / 9B / 0.8B: LoRA adapter (r=16) plus a pointer head on Qwen3-Base, no text generation at inference.
- Kev-9B trails proprietary Jev by 3.5 points on the new-source dev set (0.822 vs 0.857); test set 0.852.
- Ships trainable and runnable on a single GPU — the first credible self-hosted answer to TypeSafe's System One pitch.
- Two days on HN; part of a fast-forming Jev-alike ecosystem the browser-agent crowd is already building on.
- Sept 22: phased rollout begins for eligible API accounts — automated abuse detection runs without OpenAI staff ever seeing prompts or responses.
- Customer data can stay on customer-controlled infra or sit in OpenAI storage under customer-held encryption keys.
- Only a narrow safety signal reaches OpenAI; enterprise content is walled off from training and human review by default.
- Direct contrast with Anthropic's Enterprise Frontier Safeguards from Sept 1 — the two labs are now competing on how little they retain.
01
Anthropic Ships Claude Opus 5.5 at Fable-Level Quality for 40% Less
models anthropic.com
02
OpenAI Adds GPT-6 Sol and GPT-6 Luna, Halving API Prices
models openai.com
03
Pentagon Probe Pins Iran School Strike on Palantir Maven Overreliance
industry bloomberg.com
04
jev-ultrafast Hits 17K Stars in a Week by Deleting the Text From Web Agents
tools github.com
05
Kev Ports Jev's Decision-Only Architecture to Qwen3.5 for Local Runs
open-source github.com
06
OpenAI Turns On Private Safety Processing for API Zero-Data-Retention Customers
industry openai.com