Signals
Claude Opus 5.5 vs GPT-6 Sol: Two Launches, 100 Minutes Apart, Two Different Bets
By the Top5Apps editorial team · Published September 23, 2026 · Updated September 23, 2026 · 5 min read
Short answer: Anthropic won the afternoon on quality; OpenAI won it on price — and only one of the two put its new model where subscribers can actually use it. Claude Opus 5.5 landed Tuesday at 12:31 PM Eastern as 'the first model in our new Claude 5.5 family,' priced at $4 in / $20 out per million tokens and live on Claude's $20 Pro plan the same day. About a hundred minutes later, GPT-6 Sol and GPT-6 Luna arrived at $2/$10 and $0.10/$0.50 — half their GPT-5.6 promotional prices, and OpenAI says permanent — but 'not yet available in Chat' on any ChatGPT plan. Independent testing within 24 hours put Opus 5.5 first on every index that scored it; it put Sol level with the model it replaces.
Claude@claudeai
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family. It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
View original post ↗
OpenAI@OpenAI
Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe… We've also made caching and inference more efficient, and we're passing the savings directly to you: 50% lower API prices for Sol and Luna compared with GPT‑5.6 promotional pricing.
View original post ↗
The prices, side by side
| Model | Input / output per 1M tokens | Cached input | Positioning |
|---|---|---|---|
| Claude Fable 5.1 | $10 / $50 | $0.25 | Anthropic's flagship (Sept 1) |
| Claude Opus 5.5 | $4 / $20 | $0.20 | New: 'Fable 5.1-level on most work'; replaces Opus 5 ($5 / $25) |
| GPT-6 Astra | $10 / $50 | $1.00 | OpenAI's flagship (Sept 3) |
| GPT-6 Sol | $2 / $10 | $0.20 | New: the mid tier; was $4 / $20 as GPT-5.6 Sol (promo) |
| GPT-6 Luna | $0.10 / $0.50 | $0.01 | New: the small tier; was $0.20 / $1.20 as GPT-5.6 Luna |
Read Anthropic's '40% cheaper' carefully: it's a workload claim, not a price cut. The list price is 20% under Opus 5; the 40% comes from cheaper tokens plus fewer tokens per task, and cache reads dropped 60%. OpenAI's '50%' is a real list-price halving — against the GPT-5.6 promotional rate, which was due to expire in November anyway. Both numbers are true; neither is what the headline implies.
What the neutral scoreboards say (24 hours in)
| Claude Opus 5.5 | GPT-6 Sol | For reference | |
|---|---|---|---|
| Artificial Analysis Intelligence Index | 58 (max effort) — #1 of 212 | 48 (max) — level with GPT-5.6 Sol's 47 | Fable 5.1: 53 · GPT-6 Astra: 53 · Luna: 37 |
| Vals Index | #1 — 69.69%, $22.30 per test | #8 — 62.57%, $7.56 per test | GPT-5.6 Sol: #7 at 63.71% · Astra: #4 |
| CursorBench 4.0 (Cursor's coding eval) | 57.8% at Max — ranks 1–4 at every effort level | Not yet listed | Fable 5.1: 51.8% · Opus 5: 46.6% · GPT-5.6 Sol: 41.7% |
| Speed (AA) | 74.7 tokens/s at medium effort; 'very verbose' at max | 126 tokens/s — up from 82 for 5.6 Sol | Luna: 157 tokens/s |
Two things stand out. Opus 5.5 outperformed its own marketing — Anthropic hedged that 'the gap between Opus 5.5 and Claude Fable 5.1 is narrower than these scores suggest,' and then the independent indexes placed it above Fable 5.1, not level with it. Sol did roughly what OpenAI said and no more: Artificial Analysis's verdict was that its scores 'remain level with GPT-5.6, with progress in some evaluations and regressions in others' — notably a roughly 100-point drop on the GDPval office-work benchmark — while the cost per task halved. Sol is a price release. Opus 5.5 is a model release.
Where you can actually use them — the part that decides it for subscribers
- Claude Pro ($20): Opus 5.5 is in the app and in Claude Code, where it's now the default Opus — and Anthropic switched Pro's Claude Code default from Sonnet to Opus and raised five-hour limits on Pro, Max, and Team the same day. Fable 5.1 remains a usage-credit add-on on Pro. Free stays on Sonnet and Haiku.
- ChatGPT (every paid plan): Sol and Luna run in ChatGPT Work and Codex only — 'not available in regular ChatGPT conversations.' Chat still answers with GPT-5.6 Sol on Plus and GPT-5.6 Luna on Free and Go. Free and Go users can try GPT-6 Luna in the desktop app's Work mode. Astra is unchanged: 'GPT-6 Pro' in Chat for Pro, Business, and Enterprise; Work and Codex for Plus.
- API and clouds: both live everywhere on day one — Opus 5.5 on Bedrock, Google Cloud, and Microsoft Foundry; Sol and Luna in the OpenAI API with 1.05M-token context.
- GitHub Copilot added all three within hours; Luna reaches the $10 Pro plan, Opus 5.5 and Sol need Pro+ or above. Cursor listed Opus 5.5 at $4/$20 and, as of Wednesday, hadn't listed Sol or Luna at all.
- Usage sweeteners on both sides: Anthropic gave subscribers a rate-limit reset they can bank and spend whenever; OpenAI's Tibo loaded 'a banked reset into all accounts of our Plus, Pro and Business users' — a one-time refresh of Codex and Work usage windows.
The fine print on Opus 5.5 that Anthropic said out loud
Opus 5.5 ships with Fable 5.1's biology safeguards, and the system card is explicit about what that means: 'blocks on this classifier will fall back to Claude Opus 5.' Most cybersecurity tasks 'will be re-routed to Opus 4.8.' Anthropic's Eric Kabrams put it plainly: 'most bio-related queries fall back to Opus 5.' Labs, biotech startups, and pharma teams can apply to the Life Sciences Verification Program, announced five days before the model and currently in beta for organizations; individuals come later. The fallbacks are visible in the conversation, so you'll know when it happens — but if your work is virology or molecular design, Opus 5.5 is, for now, Opus 5.
Who this was for
If you pay $20 a month for an assistant, Tuesday was Anthropic's day without much argument: the best-scoring model in independent testing is now the everyday model on Claude Pro, while ChatGPT Plus subscribers got two new models they can't use in the tab they actually live in. Our Claude review and ChatGPT review both reflect this now.
If you build on the API, it's closer than the scoreboards suggest. Sol at $2/$10 is now the cheapest frontier-adjacent model from either lab and is 50% faster than what it replaces; Luna at ten cents per million input tokens is the sleeper of the day for classification, extraction, and anything high-volume. Opus 5.5 is the better model and it costs twice Sol's rate — a worthwhile trade for agentic coding (where the Cursor and Terminal-Bench gaps are large) and a hard sell for summarizing support tickets.
Our read: the two labs are no longer competing on the same axis. Anthropic's move was to collapse the gap between its flagship and its workhorse — Fable-class quality at a fifth of Fable's price, in the consumer app on day one. OpenAI's move was to make the GPT-6 name cheap and ubiquitous in the places developers and Work users live, while keeping Chat on last generation's models and Astra as the premium tier. One of those is a product decision a subscriber can feel this week. The other is a pricing decision a CFO will feel next quarter. Watch for Sonnet 5.5 and Haiku 5.5 'in the coming weeks' — if Anthropic brings this pricing logic down the stack, the $2/$10 tier becomes a fight too.