The short version#
Both labs now sell a three-tier lineup, and the tiers map onto each other almost perfectly. OpenAI's GPT-5.6 Sol ($5/$30 per million tokens) competes with Claude Opus 5 ($5/$25). Terra ($2.50/$15) competes with Sonnet 5 ($2/$10 intro pricing until August 31). Luna ($1/$6) undercuts everything Anthropic currently offers at the low end. Above all of that sits Anthropic's Fable 5 ($10/$50) with no direct OpenAI counterpart at a public price.
If you just want the practical answer: for high-volume cheap workloads Luna is currently unanswered, for the mid-tier the Sonnet 5 intro price wins on cost until September, and at the top the choice is between Sol's cybersecurity-leaning positioning and Opus 5's published near-frontier coding numbers. Details and caveats below.
What actually launched#
GPT-5.6 went into limited preview on June 26, 2026, restricted to about 20 US-government-vetted organizations, and reached general availability in ChatGPT, Codex and the API on July 9 (Crypto Briefing, felloai). It is a lineup, not one model:
- Sol, the flagship, aimed at advanced coding, scientific research and cybersecurity work.
- Terra, the everyday model, positioned at roughly GPT-5.5-generation performance for about half Sol's price (TechJournal).
- Luna, the fast, cheap tier for high-volume tasks.
Claude 5 arrived in three releases between June 9 and July 24: Fable 5 (frontier, safety classifiers, $10/$50), Sonnet 5 (new default, $2/$10 until August 31 then $3/$15) and Opus 5 ($5/$25, published benchmarks close behind Fable 5). Our Claude 5 family guide covers the three models and the June export-control interruption in detail.
Grok 4.5 launched in the same July window but remains in private beta at SpaceX and Tesla, so it is not yet a real option for buyers and we leave it out of the comparison tables.
Pricing side by side#
Per million tokens, as published in July 2026:
| Tier | OpenAI | Price (in/out) | Anthropic | Price (in/out) |
|---|---|---|---|---|
| Frontier | (none public) | Fable 5 | $10 / $50 | |
| Flagship | GPT-5.6 Sol | $5 / $30 | Opus 5 | $5 / $25 |
| Mid | GPT-5.6 Terra | $2.50 / $15 | Sonnet 5 | $2 / $10 (to Aug 31), then $3 / $15 |
| Budget | GPT-5.6 Luna | $1 / $6 | (none comparable) |
GPT-5.6 prices per Finout's pricing breakdown and Eden AI; Claude prices per Anthropic's announcements.
Three observations from the table:
- Identical flagship input price, cheaper Claude output. Sol and Opus 5 both charge $5 input, but Opus 5's $25 output undercuts Sol's $30. For agentic workloads that generate a lot of tokens, that 17% output gap compounds.
- The mid-tier fight is a promotion war. Sonnet 5 at $2/$10 is cheaper than Terra now, but its September price ($3/$15) is the same as Terra's on output and higher on input. Anthropic set the intro window to end (August 31) suspiciously close to the GPT-5.6 launch quarter. Decide with the September prices in mind, not the July ones.
- Luna has no Claude answer. If your workload is high-volume classification, extraction or routing, $1/$6 is a tier Anthropic currently does not serve with a Claude 5 model.
Capability positioning, with appropriate skepticism#
Neither lab's launch benchmarks have been independently replicated at the time of writing, so treat this as positioning, not measurement.
Where OpenAI points Sol: advanced coding, scientific research and cybersecurity. Notably, GPT-5.6's staged rollout (government-vetted preview first) mirrors the same dual-use caution that led Anthropic to gate Mythos 5 behind Project Glasswing.
Where Anthropic points Opus 5: published results claim it more than doubles Opus 4.8 on Frontier-Bench v0.1, lands 0.5 points behind Fable 5 on CursorBench 3.2 at half the cost, and beats Fable 5 on OSWorld 2.0 computer use at about a third of the cost (Anthropic).
The tier above: Fable 5 remains the only frontier-class model an ordinary customer can buy at a public price. Its trade-offs are the $10/$50 price and safety classifiers that route fewer than 5% of sessions to an Opus 4.8 fallback in sensitive domains (Anthropic).
Which lineup for which buyer#
| Your situation | Reasonable default | Why |
|---|---|---|
| High-volume, cost-sensitive pipelines | GPT-5.6 Luna | $1/$6 tier is currently uncontested |
| General assistant / content / mid agents | Sonnet 5 now, re-evaluate Sept 1 | Cheapest mid-tier during intro window |
| Agentic coding, automation | Opus 5, benchmark Sol against it | Cheaper output, strong published coding claims |
| Cybersecurity-adjacent work | Sol | OpenAI's explicit positioning for the tier |
| Absolute frontier reasoning | Fable 5 | Only publicly priced frontier-class option |
The honest meta-point: the two lineups are converging in structure and price, which means model choice is increasingly a workload-level decision, not a loyalty decision. Run your own evals on your own tasks; the published deltas between Sol and Opus 5 are smaller than the variance most teams see between prompt versions.
What this post does not cover#
No independent benchmarks: every capability claim above is vendor-published or from the cited coverage, and we say so where it matters. Context windows and rate limits for GPT-5.6 tiers were not consistently documented across sources, so we leave them out. Grok 4.5 is excluded because private-beta access at two companies is not a market offering. Subscription-plan pricing (ChatGPT Plus, Claude Pro) is a separate topic from the API prices compared here.
Sources#
- Crypto Briefing: OpenAI launches GPT-5.6 model family
- Finout: GPT-5.6 pricing, Sol, Terra and Luna tiers
- Eden AI: GPT-5.6 Sol benchmarks, pricing and API access
- TechJournal: GPT-5.6 explained, Sol, Terra and Luna
- felloai: ChatGPT 5.6 is here
- Anthropic: Introducing Claude Opus 5
- Anthropic: Introducing Claude Sonnet 5
- Anthropic: Claude Fable 5 and Claude Mythos 5
