Quick answer: Anthropic released Claude Sonnet 5.5 on September 28, 2026, at unchanged Sonnet 5 pricing but generating output over 30% faster and using fewer tokens per task. Six days earlier, OpenAI released GPT-6 Sol and GPT-6 Luna, cutting API prices by 50% or more versus the previous generation — Luna specifically introduces a new $0.10-per-million-input-token price floor. None of these three set a new capability ceiling; all three exist to make a frontier-adjacent model meaningfully cheaper to run in production.
Claude Sonnet 5.5: Same Price, More Efficient
Anthropic's positioning for Sonnet 5.5 is unusual in that it explicitly isn't a price change — Sonnet 5.5 ships at Sonnet 5's existing rate: $2 per million input tokens, $10 per million output tokens. The claimed savings come from efficiency, not a rate cut: Anthropic says the model generates output more than 30% faster and completes tasks using fewer tokens and tool calls, which nets out to up to 30% lower cost per task even though the per-token rate hasn't moved. On benchmarks, it scores 70.6% on Terminal-Bench 4.0 and reportedly matches Opus 5.5 on several knowledge benchmarks while using fewer tokens to get there — a real gap between "same intelligence, cheaper to reach" and "smarter model," worth keeping distinct when reading the launch coverage.
GPT-6 Sol and Luna: An Explicit Price Floor
OpenAI's September 22 release of GPT-6 Sol and GPT-6 Luna takes the opposite approach — a direct, stated price cut. Sol is priced at $2 per million input tokens and $10 per million output tokens, positioned for demanding coding and agent workflows. Luna is priced at $0.10 and $0.50 respectively, aimed at focused, high-volume work — a price floor OpenAI describes as permanent rather than promotional, and roughly half of what the prior 5.6-series models cost. Both support a 1.05-million-token context window with up to 128,000 tokens of output, matching the context-window ballpark GPT-6 Astra shipped with earlier in the month.
What Actually Changed, in a Table
| Claude Sonnet 5.5 | GPT-6 Sol | GPT-6 Luna | |
|---|---|---|---|
| Released | Sep 28, 2026 | Sep 22, 2026 | Sep 22, 2026 |
| Input price | $2 / million tokens (unchanged from Sonnet 5) | $2 / million tokens | $0.10 / million tokens |
| Output price | $10 / million tokens (unchanged) | $10 / million tokens | $0.50 / million tokens |
| Positioning | Everyday tasks, coding, polished documents | Demanding coding/agent workflows | Focused, high-volume work |
| Context window | — | 1.05M tokens, 128K output | 1.05M tokens, 128K output |
Why Three Cost-Focused Releases in One Month
This isn't a coincidence of timing so much as where the competitive pressure in the industry currently sits. With frontier capability roughly matched across the top labs for a large share of real business tasks, the differentiator that actually moves a procurement decision is increasingly the cost of running a given task at production volume, not whether a lab can claim a marginally higher benchmark score. Sonnet 5.5's efficiency framing and Luna's explicit price floor are two different routes to the same conclusion: the mid-tier model, not the flagship, is where the API pricing battle is actually being fought in late 2026.
What This Means for a Production AI Budget
If you're already running a workload on Sonnet 5 or a GPT-6-series model, the practical move is checking whether Sonnet 5.5 or Luna's pricing tier fits your actual task profile before assuming your current model choice is still the cheapest fit — a high-volume, well-scoped task that doesn't need Sol- or Astra-level capability is exactly what Luna's price floor was built to undercut. This is the same evaluation discipline we cover in our vendor evaluation checklist for enterprise AI: re-check model fit against cost periodically, rather than treating your initial model choice as permanent.
If you want a second opinion on whether your current model choice still makes sense against this month's pricing, reach out at info@digit.com.pk.