cd /news/large-language-models/claude-opus-5-5-and-gpt-6-sol-ship-h… · home topics large-language-models article
[ARTICLE · art-138210] src=thenewway.ai ↗ pub= topic=large-language-models verified=true sentiment=↑ positive

Claude Opus 5.5 and GPT-6 Sol ship hours apart, each cheaper than before

Anthropic shipped Claude Opus 5.5 and OpenAI shipped GPT-6 Sol and Luna on Tuesday less than two hours apart, each at lower per-token prices. Claude Opus 5.5 costs $4 and $20 per million input and output tokens, 20% below Opus 5, and Artificial Analysis measured it 11 points ahead of Opus 5 on Terminal-Bench 4.0 at 59.6%, ranking it #1 on its Intelligence Index. GPT-6 Sol and Luna cost $2 and $10 and $0.10 and $0.50 per million tokens respectively, 50% below GPT-5.6's promotional price, while DigitalOcean opened Managed Agents in public preview, billing CPU per second of actual use.

read3 min views1 publishedSep 23, 2026
Claude Opus 5.5 and GPT-6 Sol ship hours apart, each cheaper than before
Image: Thenewway (auto-discovered)

Anthropic shipped Claude Opus 5.5 and OpenAI shipped GPT-6 Sol and Luna on Tuesday, each cheaper per token, and Artificial Analysis has first coding numbers for both. One more story inside.

Anthropic and OpenAI both shipped new models on Tuesday, less than two hours apart, and both cut prices. Claude Opus 5.5 costs 20% less per token than Opus 5, and Artificial Analysis measured it 11 points ahead of Opus 5 on Terminal-Bench 4.0, a coding test. GPT-6 Sol and Luna cost half what GPT-5.6 did at its promotional price, and score about level with it. DigitalOcean also opened a cloud for Claude Code and Codex. It bills only for the CPU they use.

Two new models, and a new place to run agents #

Anthropic ships Claude Opus 5.5 at 20% below Opus 5 per token

On Anthropic's own coding benchmark, Terminal-Bench 4.0, Claude Opus 5.5 scores 66.4%, against 52.3% for Opus 5 and 55.8% for the pricier Fable 5.1. Anthropic hedges its own table: "the gap between Opus 5.5 and Claude Fable 5.1 is narrower than these scores suggest." Artificial Analysis, an independent benchmarking firm, measured 59.6% on its own run of that test, 11 points over Opus 5, and ranks it #1 on its Intelligence Index. It costs $4 and $20 per million input and output tokens, 20% less than Opus 5. Anthropic says it will cost 40% less at default settings on typical workloads. At max effort, Artificial Analysis found it level with Opus 5 per task, because it writes 1.6 times the output tokens. Anthropic's Boris Cherny says a couple of short prompts had it write formal proofs about the Claude Agent SDK, leading to 16 pull requests fixing bugs and race conditions. Subscribers get higher five-hour limits.

OpenAI ships GPT-6 Sol and Luna at half GPT-5.6's price

GPT-6 Sol and Luna are OpenAI's cheaper tiers under GPT-6 Astra, now $2 and $10 per million tokens for Sol and $0.10 and $0.50 for Luna, 50% below GPT-5.6's promotional price. On coding, Artificial Analysis's Coding Agent Index has Sol up 2 points on GPT-5.6 Sol, with Terminal-Bench 4.0 at 43% against 37%. Luna fell 2. OpenAI's Tibo Sottiaux calls them "a very significant improvement across the board." ChatGPT Plus, Pro and Business users got a banked usage reset.

DigitalOcean opens a pay-per-use cloud for coding agents

DigitalOcean opened Managed Agents to everyone in public preview. It hosts coding harnesses "such as Claude Code, Codex CLI, and OpenCode" in sessions that can when idle and "resume d work in ~300 milliseconds." CPU is billed per second of actual use. DigitalOcean's example: two vCPUs at 25% use for an hour "would cost $0.060 in CPU and memory charges, compared with $0.126" for the full allocation. Inference and storage cost extra.

Know someone who'd want this in their inbox? Forward it — that's how this grows. And if we got something wrong, or you think we buried the real story today, hit reply. A person reads every one.

Ones to watch: early, unverified #

  • Unreal Agent : Unreal Labs says its agent harness "delivers up to 40% cost savings compared to Codex on production workloads and coding/science benchmarks, without any negative performance impact," by handling tool calls asynchronously. The maker's claim; nobody else has published a run yet.

Also worth your time #

The New Way is written with AI. It gathers the day's stories, checks them against their sources and drafts every summary. A person decides what runs and reviews every issue before we hit send.

── more in #large-language-models 4 stories · sorted by recency
── more on @anthropic 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/claude-opus-5-5-and-…] indexed:0 read:3min 2026-09-23 ·