OpenAI Ships GPT-6.1 Sol Near Astra at One-Fifth Cost
OpenAI launched GPT-6.1 Sol at DevDay on 29 September 2026, saying the model nearly matches GPT-6 Astra at one-fifth of the standard token prices, TechCrunch reported.
PromptCrates Editorial
Staff Writer
OpenAI on Tuesday, 29 September 2026, launched GPT-6.1 Sol at its DevDay event, saying the model nearly matches GPT-6 Astra on agentic coding, computer use, and professional work while charging one-fifth the standard input and output token prices, TechCrunch reported. DevDay took place in San Francisco, according to TechCrunch’s report on OpenAI’s office-suite features. The update arrives a week after GPT-6 Sol and days after the company held GPT-6.1 Astra over internal safety shortfalls, so buyers get a cheaper workhorse instead of the delayed frontier SKU.
What OpenAI claims for GPT-6.1 Sol
OpenAI positions GPT-6.1 Sol as a meaningful jump over GPT-6 Sol on programming and debugging, document understanding, and multistep workflows. On several of those fronts, the company says performance approaches GPT-6 Astra. That framing matters because GPT-6 Astra is already on the market as the benchmark Sol is measured against, while GPT-6.1 Astra, the update that had been expected as soon as October, is not launching after safety concerns surfaced in internal testing; Sol becomes the shippable substitute with a hard cost story.
Factual accuracy is another headline metric. OpenAI says the share of responses with a factual error at low reasoning effort drops from 11.4 percent on GPT-6 Sol to 7.7 percent on GPT-6.1 Sol. Across all reasoning settings, the company claims the new model’s error rate stays within 1.9 percent of GPT-6 Astra. Those numbers are vendor-reported, so procurement teams should reproduce them on internal hard-prompt suites before rewriting routing rules.
OpenAI also says GPT-6.1 Sol is more upfront about limitations and more reliable at honoring user intent and safety constraints. In challenging evaluations, it reportedly fails less often than GPT-6 Sol at flagging broken search tools, following explicit restrictions, and avoiding unauthorized outcomes. The company claims it observed no attempts to circumvent the automated safety reviewer, consistent with GPT-6 Astra and GPT-6 Sol. That claim sits beside the same-week Astra safety hold, so operators should treat “no reviewer circumvention” as a measured eval result, not a blanket warranty for every agent harness.
Availability starts immediately for Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. OpenAI notes the model is not yet available in Chat, which keeps consumer chat on a different SKU ladder for now. Teams that live in Codex agent loops will feel the change first; chat-only workflows wait for a later surface.
How Sol fits Dots and rival mid tiers
DevDay also introduced Dots, always-on agentic avatars powered by GPT-6 Astra that OpenAI describes as remarkably capable background workers for user-defined goals. TechCrunch’s Lucas Ropek reports Dots launching Tuesday for Pro and Business Premium users in eligible markets, with messaging through Slack and Teams and text-message support coming soon. Because Dots lean on Astra while GPT-6.1 Sol is the broadly available cheap workhorse, platform architects should map which jobs run on Sol sessions versus which need a named Dot with Astra-class tools and credentials.
Microsoft’s Agent 365 security controls are already in OpenAI’s integration roadmap for specialist Dots, TechCrunch noted. That enterprise seam matters when Sol becomes the default coding brain inside ChatGPT Work: identity, tool allow-lists, and audit trails still have to match the agent layer, not only the chat model picker. Readers comparing packaging can set this beside Microsoft Copilot Home, Code, and Autopilot, where multi-model choosers already force similar routing discipline.
Rival mid-tier pressure remains intense. Anthropic’s Sonnet 5.5 arrived a day earlier with a 30 percent speed claim and Opus-class cyber safeguards. OpenAI’s answer with GPT-6.1 Sol is price compression toward Astra-like agentic scores rather than a Sonnet-style latency pitch. Buyers should decide whether token price, wall-clock speed, or cyber-safeguard parity is the binding constraint before treating either launch as a default swap.
Validation steps before changing routing
Three checks belong on any migration plan. First, rerun agentic coding and computer-use evals at the concurrency and token budgets you actually pay for; a one-fifth price cut only helps if quality holds when you spend the savings on more parallel attempts. Second, confirm ChatGPT Work and Codex tenancy settings expose GPT-6.1 Sol to the right seats while Chat remains on the prior SKU. Third, retest refusal and authorization behaviors on the same hard prompts that worried safety reviewers about Astra, because “more reliable intent honoring” is a vendor claim that must survive your tools, not only OpenAI’s harness.
Finance and security should also rewrite unit-economics sheets. If Sol approaches Astra on coding at one-fifth token price, the temptation is to remove concurrency caps that were sized for Astra-class burn. Raise caps only after logging shows tool calls stay inside policy and factual-error rates on your document suite stay near the 7.7 percent low-effort figure OpenAI advertises. Pair that ops work with broader agent-risk reading such as NIST’s agent standards gap for enterprises, which stresses that model menus alone do not create governance.
Primary reporting for this article: TechCrunch’s 29 September pieces by Aisha Malik on GPT-6.1 Sol and by Lucas Ropek on Dots, plus Ropek’s office-suite report for the San Francisco venue. Anchored facts include DevDay timing; one-fifth input and output token pricing versus standard rates; near-Astra claims on agentic coding, computer use, and professional work; factual-error drop from 11.4 percent to 7.7 percent at low reasoning effort; error rate within 1.9 percent of Astra across reasoning settings; availability in ChatGPT Work and Codex for Plus through Edu; Chat exclusion; Astra hold context; and Dots’ Pro and Business Premium launch with Slack and Teams messaging.
