🚀 Claude Opus 5.5 + GPT-6 Sol / Luna

Jev Makes AI Decisions 20x Faster

Author

The Summary AI
September 23, 2026

Sponsored by

Welcome back!

The AI model race just got cheaper on both sides. Claude Opus 5.5 buys more quality per task; GPT-6 Sol and Luna buy more tasks per dollar. Two different ways to make frontier AI cheaper. Let’s unpack…

Today’s Summary:

  • 🚀 Claude Opus 5.5 tops benchmarks

  • 🔥 GPT-6 Sol and Luna halve prices

  • ⚡ Jev skips LLMs for 20x faster AI decisions

  • 🚦 Anthropic CEO proposes pacing AI race

  • ⚖️ GPT-6 Astra for legal research

  • 👁️ Qwen multimodal matches Gemini benchmarks

  • 🛠️ 2 new tools

TOP STORY

Claude Opus 5.5 brings Fable-level performance at 40% lower cost

The Summary: Anthropic launched Claude Opus 5.5, its new top model, with performance near Fable 5.1 on most work and a claimed 40% lower running cost than Opus 5. It leads benchmarks in agentic coding, computer use, and knowledge work, and runs 30% faster than Opus 5. Early users are praising its coding quality and a more natural writing style.

Key details:

  • Ranks #1 on Artificial Analysis’s Intelligence Index, scoring 58 versus 53 for Fable 5.1 and GPT-6 Astra

  • Early users are praising cleaner writing with less jargon, with “Claude is back” becoming a common reaction among Claude power users

  • Cache reads cost $0.20/million tokens, 95% cheaper than fresh input, cutting major expenses for long coding sessions

  • Available today through Claude for Pro, Max, Team, and the API

Why it matters: Opus 5.5 looks great for serious coding and long agent work, where the quality of the finished result matters. Four of its five effort levels land on the cost-performance frontier, with Medium appearing to be the sweet spot for coding. GPT-6 Sol costs less per token, but Opus 5.5 makes a strong case when your priority is quality per task.

FROM OUR PARTNERS

Your AI Agent Follows You Everywhere

AI agents built to get work done

Skydive agents take on work for you, using the same tools your team already uses.

Talk to your agent on the web, Slack, email, iMessage, or right from your terminal. Wherever you pick up the conversation, your agent keeps the context.

Put agents to work across customer support, sales, marketing, engineering, ops, and more.

Give your agent a job. They’ll take it from there.

OPENAI

OpenAI releases GPT-6 Sol and Luna at half the price

The Summary: OpenAI launched GPT-6 Sol and Luna, with API prices about 50% below GPT-5.6. Independent tests show the main gain is cost, with Luna dropping to $0.10/$0.50 per million tokens. Both models are available in Codex, ChatGPT Work, and the API, but not in regular Chat.

Key details:

  • GPT-6 Sol intelligence score barely moved from GPT-5.6 Sol, but cost per completed task fell by about 50%; Luna’s fell by about 60%

  • Early users praise Luna as a cheap workhorse, while Sol gets a more divided response, with some falling back to GPT-5.6 Sol

  • Available now in ChatGPT Work and Codex for Plus, Pro, Business and API. Not available in regular Chat

Why it matters: GPT-6 Sol and Luna push down the cost of running capable agents, without a particular jump in intelligence. Sol 6 costs half as much as Sol 5.6, but its quality gains look modest and uneven. Luna is the far more interesting product, because it now makes parallel agents and routine coding extremely cheap.

TYPESAFE AI

Jev skips chat for AI decisions

The Summary: TypeSafe AI, founded by a former OpenAI researcher, launched Jev, a new “System One” AI model built to make fast, structured decisions instead of generating text. It takes text or JSON as input and returns choices, scores, and probabilities that software can act on directly. The model targets classification use cases where using a full LLM can be slow and expensive. Early developers like that Jev sits beside frontier models, instead of trying to replace them.

Key details:

  • Jev responds in 70-500ms and costs $0.042 per million input tokens, with no output cost

  • Using Jev instead of an LLM for classification can accelerate some workflows by 20x with 20x cost savings

  • Jev can still make wrong decisions, but its probabilities let software decide when to escalate a task if it has low confidence

  • Available through TypeSafe’s API and partner services

Why it matters: AI applications often waste expensive reasoning tokens on very simple decisions. Jev can handle those decisions in milliseconds, leaving costlier LLMs for the work that needs deeper reasoning or text generation. This can accelerate agents and reduce their costs in a compounding way, as one cheap decision can prevent several unnecessary frontier-model calls.

TOOLS

🥇 New tools

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/