The clearest OpenClaw trend on Thursday, September 24, 2026 is that the model picker in the Web UI is no longer a single button. OpenClaw v2026.9.6 ships first-class chat-model support for Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7 in the same week all three vendors dropped flagship API prices. Today’s earlier cost-visibility coverage mapped the spend side and the Decision Models piece mapped the routing side; this piece maps the new model lineup.
The September cadence on this site has been tracing the same shift. The parallel cloud sessions piece mapped how a solo operator hands work to a remote worker. The specialist teams and Atomic Updates piece mapped how a single operator stands up a recoverable crew. The shared Browser pages and cold-storage archives piece mapped how a human and an agent look at the same screen. What 2026.9.6 adds is a lineup of model choices priced to be swapped mid-workflow.
Four vendors shipped flagships in roughly 48 hours
SpaceXAI’s Grok 4.7 went out first on September 21, 2026, billed as “twice as fast, at half the price” of Grok 4 with 500,000-token context for coding and knowledge work. Anthropic’s Claude Opus 5.5 followed on September 22, 2026 at $4 input and $20 output per million tokens, with cache reads at $0.20 per million, 60% lower than its predecessor. OpenAI’s GPT-6 Sol and GPT-6 Luna arrived within an hour, with Sol at $2/$10 and Luna at $0.10/$0.50 per million. VentureBeat confirms the numbers and notes Luna is 97.5% below Opus 5.5 on raw pricing.
The timing matters because 2026.9.6 ships the same week. The release notes list Claude Opus 5.5, GPT-6 Sol and Luna, and Grok 4.7 as chat-model choices available through the Gateway. A small business that paid for one model last month now has four first-party flagships to route between. Simon Willison’s September 22, 2026 post called the same window “a new price war” and emphasized that the move matters for cost-per-finished-task.
What a one-person operation actually does with four flagships
The practical question is which model goes in which slot. OpenAI’s launch page describes Sol as a step up across the family and Luna as a cheaper sister tier; Anthropic’s launch page emphasizes Opus 5.5’s agentic coding benchmarks and lower compute per task; SpaceXAI’s announcement highlights long context. The September 23, 2026 Neuron Daily live-test post makes the sharper point: the real story is cost.
A solo founder can wire the lineup in three places this week. The founder daily-ops knowledge page already shows how to keep a default chat model and override it per agent; with 2026.9.6, the override has four new choices to plug in. The pattern is to keep Opus 5.5 on the high-stakes work where the agent writes code or drafts a client-facing brief; Sol on everyday reasoning where the agent summarizes a transcript or plans a sprint; Luna on bulk work where the agent triages a hundred inbox items or extracts structured data; and Grok 4.7 on long-context work across a code base or transcript library.
The cost-controlled specialist stacks piece argued that small teams split workflows into specialist workers; with 2026.9.6, those specialists can each pick a different flagship without leaving the Gateway. The August 13 cost-routing piece argued the chat model should be reserved for the work it is uniquely good at; with a lineup this cheap, that filter runs at the slot level.
Cost per finished task, not per million tokens
Both Anthropic and the independent press pushed the same framing. The Opus 5.5 page reports a 40% lower cost on typical workloads at default settings, not just a 20% lower per-token price, because the model uses fewer tokens to finish the same task. VentureBeat makes the point bluntly: the competition is moving toward cost per completed task. Willison tracks the same metric.
For an SMB that pays per token, the practical effect is that a one-person operation can now measure spend in dollars-per-finished-task instead of dollars-per-thousand-tokens. The 30-day Usage view shipped in the same 2026.9.6 release shows the token bill; the model lineup shows where the work was routed. The Doctor CLI now warns about sessions whose usage-cost cache refresh failed, pairing with that view so a small business knows what was spent and on which model.
VentureBeat reports Sol at half of Opus 5.5’s base rates and Luna at roughly 2.5% on raw input and output. That floor turns a use case that was uneconomic last month — triaging a thousand community comments or summarizing a year of customer emails — into something a one-person operation can pay for. The September 22 Towards AI coverage flags the same idea: Grok 4.7 is the long-context budget flagship, useful where Opus 5.5 is overkill and Luna is too cheap to trust.
Remote workspaces now carry Files, Memory, and Skills
The second v2026.9.6 piece that matters for solo operators is what remote workspaces gained. The release notes call out that “remote workspaces gain Files, Memory, and Skills,” with the official @openclaw account highlighting the same line. Superpower Daily’s recap restates the point. For a one-person business on a cloud workspace, the agent can now read the same files, see the same memory, and call the same skills it would on the local install.
For a creator running a newsletter on a remote workspace, drafts and brand assets now travel with the agent. The cron jobs knowledge page shows how to wire scheduled tasks; with remote workspaces carrying the same skill set, a scheduled job can land on the laptop or the cloud and produce the same answer.
Limits and what to watch next
Two limits are worth flagging. First, capability is not identical across the four flagships. Opus 5.5 beats the previous general-access flagship on most reported benchmarks; the Towards AI note flags that Grok 4.7 headline numbers require the right effort level; Luna and Sol are cheaper because they trade capability for throughput. Second, the macOS app for 2026.9.6 can crash on every launch after the in-app update, with a documented issue (#156861) tracked on GitHub; the macOS build has been withdrawn from the Sparkle feed while a 2026.9.7 hotfix is in progress. A macOS user should reinstall the 2026.9.5 macOS build until then.
The trend line is clear. The Gateway now exposes four first-party flagships priced within an order of magnitude, plus a Decision model slot for small routing questions. A one-person business that needs Opus-class reasoning for one workflow and Luna-class throughput for the next can wire both from the same Web UI this week. The next release is likely to keep closing that loop: per-session cost reporting that names the model, default Decision-model routing that picks the cheapest qualifying chat model, and remote workspaces that mirror the laptop’s skill set.
For a solo founder, creator, or SMB operator in late September 2026, the operating system is still on the laptop. What 2026.9.6 changes is that the choice is no longer which single model to commit to; it is which slot gets which flagship this morning, with the option to swap before lunch without leaving the Gateway.