The Frontier Goes Cheap: Anthropic and OpenAI Fire Competing Price Cuts 90 Minutes Apart
The clearest signal of the week came on Sept 22 (US time): two of the three frontier labs dropped lower-cost models within 90 minutes of each other.
Anthropic — Claude Opus 5.5. The first model in a new 5.5 family, Opus 5.5 is priced at $4 / $20 per million tokens — a 20% cut versus Opus 5 — and Anthropic says it costs ~40% less to run on typical workloads while matching Fable 5.1 on most tasks (Artificial Analysis Intelligence Index 58, the highest the index has measured). It ships with the same cyber/bio/frontier safeguards previously reserved for Fable, plus external testing from METR and Frontier Design. Sonnet 5.5 and Haiku 5.5 follow in coming weeks.
OpenAI — GPT-6 Sol and GPT-6 Luna. Sol lands at $2 / $10 per million tokens and Luna at $0.10 / $0.50 — exactly half the GPT-5.6 promotional rates, billed as standard (not limited-time) pricing. Both inherit much of GPT-6 Astra's capability through improved caching and inference efficiency; Sol enters the value tier as OpenAI's best-cost-per-task option. Both go live in ChatGPT Work and Codex from Thursday.
The read: open-weight pressure (DeepSeek, Qwen, Kimi, MiMo) is now doing more to set frontier pricing than any safety pledge. The race is shifting from "who is strongest" to "who deploys near-frontier capability cheapest, fastest, and at scale."
Google Ships Production-Grade Voice: Gemini 3.8 Flash TTS and Flash-Lite TTS
Google DeepMind added two text-to-speech models to the Gemini API and AI Studio: Gemini 3.8 Flash TTS and the cheaper Flash-Lite TTS, both with promptable voice design.
Highlights: 2,000+ production voices, 30-second consented voice cloning protected with SynthID and C2PA provenance, 100+ languages and dialects, hours-long two-speaker scenes, and markup for vocal bursts. Flash ranks #1 on Hume Voice Design (71.4) and accent (60.8); Flash-Lite takes #2 on Hume Overall Quality. No public launch pricing was listed.
This is the same pattern as the live-audio push two weeks ago — Google is treating speech as a commodity modality: high fidelity, low friction, watermarked by default. The competitive moat is now distribution (Chrome, Android, Workspace) more than raw quality.
Claude Ran an Autonomous Science Sweep and Found a New Enzyme System
Anthropic's new life-sciences research group used Claude to autonomously search a massive DNA sequence database and surface a previously unknown enzyme system: array-associated reverse transcriptases (ART).
The run used roughly 950 parallel agents over 21 hours and 210 million tokens, identifying an unusual reverse transcriptase in bacteriophages accompanied by a CRISPR-like repeat array and an accessory protein. MIT's Feng Zhang called the finding "genuinely intriguing." A preprint is out; function is still under investigation.
It's a concrete data point for the "AI as a science-discovery engine" thesis: not just summarizing papers, but designing and executing a large parallel search across biological sequence space and surfacing candidates humans hadn't categorized. Expect more labs to staff "agent swarms" for discovery, not just coding.
Daily AI briefing from AI Nexus Daily.
Top comments (0)