# Opus 5.5 and GPT-6 Sol shipped the same day, and neither launch table could see the other

URL: https://www.thedeepfeed.ai/posts/2026-09-23-opus-5-5-and-gpt-6-sol-same-day/
Category: Models
Published: 2026-09-23
Author: the-deep-feed
Tags: anthropic, openai, claude, gpt-6, pricing, benchmarks
Kind: news

> Anthropic released Claude Opus 5.5 at $4 and $20 per million tokens and OpenAI released GPT-6 Sol at $2 and $10 about 90 minutes apart on September 22, and because each launch table benchmarks the other lab's previous model, the price sheets are the only like-for-like comparison.

## TL;DR

- **Claude Opus 5.5** lists at **$4** and **$20** per million input and output tokens, 40% of Fable 5.1's rate, and Anthropic says it performs at Fable 5.1's level on most work.
- **GPT-6 Sol** lists at **$2** and **$10** and GPT-6 Luna at **$0.10** and **$0.50**, a 50% cut from GPT-5.6 promotional pricing; Sol and Opus 5.5 both charge **$0.20** per million cached input tokens.
- Neither launch table includes the other model: Anthropic compares Opus 5.5 with **GPT-6 Astra** and GPT-5.6 Sol, OpenAI compares Sol with Claude Opus 5 and Fable 5.1, and every score on both pages is vendor-reported.

On September 22, **Anthropic** released [Claude Opus 5.5](https://www.anthropic.com/claude-opus-5-5) at $4 per million input tokens and $20 per million output. **OpenAI** released [GPT-6 Sol and GPT-6 Luna](https://openai.com/index/introducing-gpt-6-sol-and-luna/) the same day, with Sol at $2 and $10. The two announcements reached Hacker News 91 minutes apart, [Anthropic's](https://news.ycombinator.com/item?id=49803892) at 16:29 UTC and [OpenAI's](https://news.ycombinator.com/item?id=49805509) at 18:00. Each lab published a benchmark table with its launch, and neither table contains the model the other lab launched. Until someone runs both, the only like-for-like comparison is the price sheet, and on the price sheet the two companies are now selling the same thing: near-flagship work at a fraction of flagship cost per task.

# Two price sheets, one cache rate

Anthropic's pitch is that Opus 5.5 "performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5." The list price supports the second half directly. Opus 5.5 costs 20% less per token than Opus 5's $5 and $25, and its cache reads fall 60%, to $0.20. Fable 5.1, which Anthropic [released on September 1](https://www.anthropic.com/claude-fable-and-mythos-5-1) alongside the restricted Mythos 5.1 ([covered here](/posts/2026-09-02-claude-fable-and-mythos-5-1/)), lists at $10 and $50, so Opus 5.5 sells Fable-class work, by Anthropic's account, at 40% of Fable's per-token price.

OpenAI's pitch is a straight halving. Its post says it is "reducing API prices for Sol and Luna by 50% compared with their GPT‑5.6 promotional pricing," taking Sol from $4 and $20 to $2 and $10 and Luna from $0.20 and $1.20 to $0.10 and $0.50. The [GPT-6 Sol model page](https://developers.openai.com/api/docs/models/gpt-6-sol) lists cached input at $0.20 per million tokens.

| Model (list price per 1M tokens) | Input | Output | Cached input |
|---|---|---|---|
| GPT-6 Luna | $0.10 | $0.50 | n/a |
| GPT-6 Sol | $2 | $10 | $0.20 |
| Claude Opus 5.5 | $4 | $20 | $0.20 |
| Claude Fable 5.1 | $10 | $50 | $0.25 |
| GPT-6 Astra | $10 | $50 | $1.00 |

The cache column is where the two launches meet. For coding agents that resend a repository's context on every turn, cache reads are most of the bill, as Anthropic's own post puts it: cache reads "make up the majority of agentic and coding work costs." On that line, Anthropic's mid-priced flagship and OpenAI's mid-tier now charge the same $0.20. On fresh tokens Sol is half the price of Opus 5.5, and both sit far below [GPT-6 Astra, which OpenAI launched on September 3](/posts/2026-09-04-gpt-6-astra-launch/) at $10 and $50.

Each lab also sold speed and access. Opus 5.5 has a Fast mode at $8 and $40 with "up to 2.5x speed," Anthropic says it generates output "more than 30% faster than Opus 5," and it raised five-hour usage limits on its Pro, Max, Team and seat-based Enterprise plans. Opus 5.5 is generally available on Anthropic's platform, AWS, Google Cloud and Microsoft Azure. Sol and Luna reached ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, with Luna in the desktop app for Free and Go users, but "These models are not yet available in Chat."

# Benchmarks aimed at last month's rival

The tables were written before either lab could test the other's new model, so each benchmarks against what was on the market the day before. All of the following figures are vendor-reported.

Anthropic's table sets Opus 5.5 against GPT-6 Astra and GPT-5.6 Sol. Opus 5.5 leads Astra on Terminal-Bench 4.0, 66.4% to 57.9%, on FrontierCode v1.1 (Main), 54.4% to 53.3%, and on Humanity's Last Exam with tools, 67.7% to 57.2%. Astra keeps two rows: AutomationBench, 41.4% to 40.0%, and Terminal-Bench Science, 64.6% to 58.7%. The footnotes matter. The Terminal-Bench comparison uses "Claude Opus 5.5 at xhigh effort and GPT-6 Astra at high effort, as reported by OpenAI," each model's best score, with a standard error of ±2.6 points for Opus 5.5. Anthropic also ran Opus 5.5 "with its production safeguards enabled," so that when they intervened, cybersecurity tasks went to Claude Opus 4.8, which it says "likely reduces Claude Opus 5.5's performance on these benchmarks."

OpenAI's post sets GPT-6 Sol against Claude Opus 5 and Fable 5.1. On AutomationBench, Sol at xhigh effort scores 33.2% at $0.27 a task, against 26.9% for Opus 5 at 11.1 times the cost and 31.4% for Fable 5.1, whose cost OpenAI says "omits the cost of the Opus 5 fallbacks, which occurred on ~40% of tasks." On DeepSWE v1.1, Sol at maximum effort reaches 68.8%, "within 1.1 percentage points of Claude Fable 5's highest score," at about 80% lower cost per task. OpenAI notes that "Evaluations of competitor models were taken from publicly available reports," and on September 22 none existed for Opus 5.5.

AutomationBench is the one row both tables share, and both print the same reference points: Opus 5 at 26.9% and Fable 5.1 at 31.4%. Against those anchors, Anthropic reports Opus 5.5 at 40.0% and OpenAI reports Sol at 33.2%. The runs differ, though. Zapier ran Opus 5.5 during early access with safeguard interventions scored as failures, and OpenAI ran Sol on AutomationBench 1.0.6 at a stated effort level. The two numbers sit on the same scale without being a head-to-head result.

Readers noticed the gap within hours. On the Opus thread, [one commenter](https://news.ycombinator.com/item?id=49808135) wrote: "So funny how both OpenAI and Anthropic post outdated pages at the same time. Opus 5.5 has benchmarks against Sol 5.6, and Sol & Luna 6 have their benchmarks against Opus 5." On the Sol thread, [another](https://news.ycombinator.com/item?id=49805949) made the point both launch posts imply: "cost per token is not as helpful anymore as cost / task."

# Cost per task is the shared pitch

Both companies argue in the same unit. Anthropic says Opus 5.5 at default effort scores 54.6% on FrontierCode, "beating GPT-6 Astra's top score (53.3%) for about a fifth of the cost per task," and that on Terminal-Bench 4.0 "It matches GPT-6 Astra at about 40% of the cost." OpenAI says Sol beats Opus 5 on AutomationBench "at just 9% of Opus 5's cost per task." Neither claim can yet be checked against the other company's new model, but the agreement on the metric is the clearer signal. Per-token list prices now understate the competition, because the labs are pricing the tokens a task consumes as much as the rate per token.

The products still differ in what a developer can switch off. The Sol model page lists reasoning effort from `none` to `max`. Anthropic says Opus 5.5 "is also no longer available with 'thinking' mode switched off," ships with the "preserved thinking" anti-distillation safeguard introduced with Fable 5.1, and routes most cybersecurity tasks to Opus 4.8 because the model "is comparable to Claude Mythos 5.1 in biology and cybersecurity." On launch day, [one developer on the Opus thread](https://news.ycombinator.com/item?id=49808591) reported that "Quite literally not a single task was able to complete today without getting flagged for [cyber]." OpenAI's Sol post gives alignment one section and makes no comparable routing announcement.

Anthropic framed Opus 5.5 as "our first release since we called for pacing the frontier," and it shipped on the same Tuesday as its largest competitor's mid-tier refresh. In July, Opus 5 [set the near-frontier price at half of Fable 5](/posts/2026-07-24-half-the-price-of-frontier/). Two months later, Anthropic's near-frontier model costs 40% of Fable 5.1, OpenAI's mid-tier costs a fifth of Astra, and both bill cached context at $0.20. As of September 23, neither lab had published a table that puts Opus 5.5 and GPT-6 Sol side by side. Buyers who want that comparison will have to run it themselves or wait for someone else to.

## Sources

- [Anthropic — Introducing Claude Opus 5.5 (Sep 22, 2026)](https://www.anthropic.com/claude-opus-5-5)
- [OpenAI — Introducing GPT-6 Sol and Luna (Sep 22, 2026)](https://openai.com/index/introducing-gpt-6-sol-and-luna/)
- [OpenAI API docs — GPT-6 Sol model page (pricing, effort, cutoff; Sep 23, 2026 snapshot)](https://developers.openai.com/api/docs/models/gpt-6-sol)
- [Anthropic — Claude Fable 5.1 and Claude Mythos 5.1 (Sep 1, 2026)](https://www.anthropic.com/claude-fable-and-mythos-5-1)
- [Hacker News — Claude Opus 5.5 launch thread (Sep 22, 2026)](https://news.ycombinator.com/item?id=49803892)
- [Hacker News — GPT-6 Sol and Luna launch thread (Sep 22, 2026)](https://news.ycombinator.com/item?id=49805509)

---

Canonical: https://www.thedeepfeed.ai/posts/2026-09-23-opus-5-5-and-gpt-6-sol-same-day/
Site: https://www.thedeepfeed.ai
Full corpus: https://www.thedeepfeed.ai/llms-full.txt