OpenAI Unveils GPT-6.1 Sol Bringing Near-Astra Intelligence at Fifth of Token Price

OpenAI has officially launched GPT-6.1 Sol, a new iteration designed to bridge the gap between cost efficiency and high-end AI performance. Announced at DevDay 2026, the model targets agentic coding, computer use, and professional workflows by achieving performance levels close to OpenAI's flagship GPT-6 Astra at one-fifth of Astra's standard token prices. Standard API pricing is set at $2 per million input tokens and $10 per million output tokens, while cached input tokens cost $0.10 per million.

The release follows industry shifts where economic efficiency in multi-step agentic execution has become a central battleground for developers and enterprises. OpenAI positions the model to handle demanding workloads where repeated reasoning calls make flagship models economically restrictive.

Benchmark Performance Across Professional Tasks

GPT-6.1 Sol delivers substantial capability gains over its predecessor, GPT-6 Sol, across software development, document parsing, and multi-step business automation. On DeepSWE v1.1, which tests AI agents on original software-engineering tasks within real codebases, GPT-6.1 Sol matches GPT-6 Astra's performance at roughly one-fifth of the cost, while outperforming GPT-6 Sol by 6.4 percentage points at a lower reasoning effort level.

In professional document comprehension via the GDP.pdf benchmark, the model scores higher than competing systems such as Opus 5.5 with fallbacks while operating at less than half the cost per task. On AutomationBench, which evaluates business workflows across 47 enterprise tools, Sol scores 2.2 percentage points above Opus 5.5 at medium reasoning effort, representing a 4.8 percentage point increase over GPT-6 Sol under identical test conditions.

Computer Use and Scientific Research Capabilities

For tasks requiring interaction with graphical software applications, GPT-6.1 Sol demonstrates measurable progress on OSWorld 2.0. At maximum reasoning effort on the offline evaluation set, it outperforms GPT-6 Sol by seven percentage points while coming within 2.1 percentage points of Astra's score at approximately one-seventh of Astra's cost per task.

In scientific workflows evaluated via Terminal-Bench Science 0.1, GPT-6.1 Sol more than doubles GPT-6 Sol's score at maximum reasoning effort. It achieves this at an average cost of $5.47 per task, compared with $23.21 for Opus 5.5 and $23.80 for Astra, delivering comparable capabilities at a cost reduction exceeding 75 percent. OpenAI notes that Astra retains the highest score among tested models at 68.1% and remains recommended for the most demanding scientific research workloads.

Caching Architecture and API Deployment Economics

Alongside lower token costs, OpenAI has restructured prompt caching for GPT-6 to improve cache hit rates and reduce redundant processing across billions of daily requests. Cached input is priced at $0.10 per million tokens, representing a 95% discount compared with standard input pricing and a 50% reduction from GPT-6 Sol's cached tier.

Developers can manage these efficiencies through the Prompt Caching Dashboard and diagnostics tools. Furthermore, adjustments to reasoning effort and tool availability now preserve existing context for cache reuse. Explicit prefix breakpoints give developers granular control over cache retention during long conversations and multi-step agentic loops.

Safety Alignment and Ecosystem Availability

OpenAI's system card addendum indicates that GPT-6.1 Sol achieves improved alignment scores compared with GPT-6 Sol, showing reduced failure rates regarding transparency over broken tools and respect for explicit constraints. Under OpenAI's Preparedness Framework, the model is rated as Critical in cybersecurity and High for biological and chemical capabilities.

According to DataCamp, unlike GPT-6 Sol released on September 22, GPT-6.1 Sol incorporates instruction-following reasoning traces, halves cached input pricing to $0.10 per 1M tokens, and enforces Critical-tier cybersecurity guardrails. According to OpenAI Deployment Safety, openAI's system card addendum classifies GPT-6.1 Sol as Critical in cybersecurity and High in biological/chemical capability, requiring full Astra-tier safeguard protocols and phased access for advanced cyber workflows.

GPT-6.1 Sol is available starting immediately within ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, though it is not yet deployed in standard ChatGPT. Developer access is provided through the OpenAI API under the model identifier `gpt-6.1-sol`, with an Ultrafast variant offering up to eight times faster token generation slated for release in the coming days.

Reading the price-performance claim correctly

A lower price per million tokens does not automatically mean a lower total bill for every agent workflow. Developers also pay for repeated reasoning steps, tool calls, retries and the larger prompts created as a task unfolds. The relevant comparison is cost per correctly completed task on an identical workload, including failed runs, rather than price per input token alone. The reported benchmarks provide useful directional evidence but may not match a company's internal documents or applications. Before migrating, teams should check API compatibility, caching behaviour, safety controls and whether their required reasoning and tool-calling features are available.

Reference:

OpenAI DevDay 2026 recap.