OpenAI and Anthropic have announced new AI models, and OpenAI has begun rolling out a product for ongoing agent work. For Canadian service businesses, the useful question is whether any of these changes can improve a specific task at an acceptable total cost. This September 30 edition covers announcements dated September 28–29; the edition date does not mean every story happened today.
OpenAI introduces GPT-6.1 Sol
Announcement date: September 29, 2026.
OpenAI introduced GPT-6.1 Sol as an upgrade to GPT-6 Sol. The company reports improvements in coding, document work and tasks that require using computer applications. Its announcement says the model is available through the API and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. At the announcement, it was not yet available in Chat. These are product availability statements and vendor evaluation results, rather than Inquory's own tests. Source: OpenAI's GPT-6.1 Sol announcement.
Inquory's interpretation: a stronger model is worth comparing on work your business already understands. One example is turning a fictional job enquiry into a draft intake record while identifying missing information. A convincing paragraph is only part of the result: the phone number, requested service and unresolved questions must also be correct. Do not change a working process solely because a vendor reports a higher benchmark score.
Anthropic launches Claude Sonnet 5.5
Announcement date: September 28, 2026.
Anthropic describes Sonnet 5.5 as an upgrade for everyday tasks, bug fixing and producing documents, slides and spreadsheets. It reports output generation more than 30% faster than Sonnet 5 and up to 30% lower cost per task in its testing. The announcement also says the input and output token prices are unchanged from Sonnet 5. Lower cost per task is therefore a vendor claim about work performed, not a claim that every API token is cheaper. Anthropic says the model is available on the Claude Platform and through AWS, Google Cloud and Microsoft Azure. Source: Anthropic's Sonnet 5.5 announcement.
Inquory's interpretation: compare the final usable result, including correction time. A fast first draft that needs substantial repair may be less useful than a slower draft that follows your instructions. Keep the task, input and review criteria the same when comparing models. A rule-based template may still be sufficient for a routine reminder or a message assembled from fixed fields.
OpenAI begins rolling out dots
Announcement date: September 29, 2026.
OpenAI announced dots, agents that use their own cloud computer and connected applications to work on ongoing tasks. The company says rollout is beginning for Pro and Business Premium users in eligible markets; Enterprise users, including Edu and Healthcare, can try a beta when their administrator enables it. Its announcement describes app permission controls, rules for actions and automated review of consequential actions. It also warns that dots can make mistakes and that consequential work should be reviewed. Availability is limited by rollout, plan and market. Source: OpenAI's dots announcement.
Inquory's interpretation: the first business decision is how much access a task needs. Preparing a draft follow-up is different from sending it to a customer; suggesting a correction is different from writing it to a CRM. For a first trial, choose a draft-only task and inspect the result before allowing an action that changes records, sends messages or makes commitments. The announcement does not establish reliability for your particular systems.
A practical way to evaluate the news
This is a proposed comparison worksheet, not a reported Inquory or customer experiment. Use fictional or appropriately sanitized examples first. Before testing, decide what counts as an acceptable result and which errors would stop the trial.
| Question | What to record |
|---|---|
| Did it preserve the important facts? | Missing, altered or invented details in each output |
| Could someone use the result? | Corrections required and time spent reviewing |
| Did it respect the boundary? | Any attempt to send, modify or access more than instructed |
| What did completion cost? | Measured model charges, retries and review time |
Keep the comparison small enough to inspect every result. If a task only copies fixed fields or applies a clear condition, try simple rules first. If it needs interpretation, use a repeatable set of examples and preserve the failed cases. Start with the AI pilot measurement guide for a baseline and stop conditions, or the lead qualification guardrails for intake boundaries. The ROI calculator can support an illustrative calculation once your inputs are explicit.
About this edition
Inquory Research prepared this brief with AI assistance and checked the linked primary announcements on September 30, 2026. We have not independently benchmarked these products, verified availability in every Canadian account, or measured customer savings. The practical interpretations and worksheet are Inquory's proposed methods. This is a selected brief, not a comprehensive news feed. See our editorial method, AI-use disclosure and corrections policy.