OpenAI Launches GPT-6.1 Sol: Near Astra-Level Power at Lower Cost

TL;DR

• OpenAI has launched GPT-6.1 Sol, a mid-tier upgrade that delivers nearly 98% of GPT-6 Astra's performance on key benchmarks at roughly half the cost.
• The model brings major gains over GPT-6 Sol in code generation and debugging, long-document understanding up to 1M tokens, and reliable multi-step business workflows.
• For developers and enterprises, GPT-6.1 Sol is positioned as the new default workhorse, with same-day API access and significantly lower pricing for input and output tokens.

Astra Power Without the Astra Price Tag

OpenAI has officially unveiled GPT-6.1 Sol, the latest evolution in its GPT-6 family, and the pitch is simple: flagship-level intelligence without flagship-level bills.

Announced this week, GPT-6.1 Sol sits directly between the standard GPT-6 Sol and the top-tier GPT-6 Astra. According to OpenAI, it matches Astra within 1-3% on general reasoning, math, and instruction-following benchmarks, while costing developers substantially less to run in production. Early pricing shared by OpenAI shows input and output tokens priced around 40-50% lower than Astra, making it an immediate contender for high-volume applications.

The launch continues OpenAI's split-model strategy, with Astra reserved for frontier research and the most complex reasoning tasks, while Sol handles efficient, scalable deployment.

What Changed Since GPT-6 Sol

While the version bump from 6.0 to 6.1 sounds incremental, OpenAI says the under-the-hood improvements are significant. GPT-6.1 Sol was trained with an upgraded post-training stack focused on reliability, tool use, and reduced hallucinations.

OpenAI reports a 34% drop in factual errors on long-form responses compared to GPT-6 Sol, plus faster response times and a 22% improvement in token efficiency. That means shorter, more precise answers that cost less to generate. The model is available starting today via the API and ChatGPT for Plus, Pro, and Enterprise users, with default routing for many coding and business use cases.

Code Writing and Debugging Get a Major Boost

The biggest leap is in software development. OpenAI claims GPT-6.1 Sol scores nearly on par with Astra on SWE-Bench Verified and its internal debugging evaluations, up more than 18 points from GPT-6 Sol.

In practice, that translates to cleaner first-pass code in Python, JavaScript, TypeScript, Rust, and Go, better handling of large codebases, and far more accurate bug detection across multiple files. The model is also significantly better at explaining its fixes, using terminal tools, and running iterative test-and-patch loops without human guidance.

Developers in the early access program report that GPT-6.1 Sol can now refactor legacy code, migrate frameworks, and maintain context across entire repositories with minimal drift, tasks that previously required stepping up to Astra.

Built for Documents and Long Context

Document understanding is the second pillar of the upgrade. GPT-6.1 Sol supports up to a 1-million-token context window in the API, double that of the original Sol, with much stronger recall at the middle and end of long inputs.

OpenAI says the model achieves 96% of Astra's accuracy on finance, legal, and scientific document QA tests. That includes summarizing 500-page PDFs, comparing contract versions, extracting tables from earnings reports, and answering multi-hop questions across thousands of pages.

Crucially, citation accuracy and source grounding have improved, addressing one of the biggest complaints about GPT-6 Sol in enterprise search and knowledge management deployments.

A Workhorse for Multi-Step Business Workflows

The third major focus is agentic, multi-step workflows. GPT-6.1 Sol was rebuilt for native tool calling, web browsing, function execution, and coordination across business apps like Salesforce, Slack, Excel, and GitHub.

OpenAI says it completes 5- to 15-step tasks, such as generating a sales report, reconciling invoices, booking travel, or triaging support tickets, with a 41% higher success rate than its predecessor. It also holds state better, asks fewer clarifying questions, and recovers more gracefully when a tool fails.

That reliability push is paired with new enterprise controls, including longer memory for projects, improved permission handling, and audit trails for agent actions.

What It Means for Developers and Enterprises

For developers, GPT-6.1 Sol is likely to become the new default. It offers Astra-class reasoning for most everyday coding, chat, and automation tasks, with lower latency and pricing that makes scaling far more viable. OpenAI is encouraging teams to downgrade from Astra to 6.1 Sol unless they need frontier math, deep scientific reasoning, or ultra-complex planning.

For enterprises, the message is about ROI. Lower cost per query, higher accuracy on documents, and more dependable agents mean customer support bots, internal copilots, and back-office automations can run at scale without constant human oversight. CIOs evaluating AI budgets will see GPT-6.1 Sol as a way to expand deployment while cutting inference spend.

With competitors like Google Gemini 3.5, Anthropic Claude 4.5, and open-weight models closing the gap, OpenAI is betting that near-flagship performance at mid-tier pricing will keep Sol at the center of the AI stack.

AndroGuider Team

Articles written by the AndroGuider team. We try to make them thorough and informational while being easy to read.