Key Notes
- GPT-6.1 Sol improves coding, document and agent performance a week after the previous Sol launch.
- Standard API input and output cost one-fifth of Astra rates, while cached input falls to $0.10 per million tokens.
- Access starts in ChatGPT Work and Codex for eligible plans, with regular Chat still pending.
OpenAI has released GPT-6.1 Sol, a week after the arrival of GPT-6 Sol, bringing its more affordable model closer to flagship Astra performance. The September 29 DevDay release targets coding, document analysis and work carried out by AI agents.
The central change is capability at the existing Sol price point. Standard input and output tokens still cost one-fifth as much as Astra’s, while cached input becomes cheaper. OpenAI’s results suggest a narrower performance gap on several tasks, although the company continues to position Astra for its most demanding work.
Coding and Business Workflows Improve
In the announcement, OpenAI reports that GPT-6.1 Sol matches Astra on DeepSWE 1.1 at roughly one-fifth the cost. It beats the previous Sol’s best result by 6.4 percentage points while using a lower reasoning setting and costing less.
The company also reports near-Astra results on GDP.pdf, which tests questions about complex professional documents. On AutomationBench, the new model improves on GPT-6 Sol by 4.8 percentage points at medium reasoning effort. On OSWorld 2.0’s offline computer-use test, it finishes within 2.1 points of Astra at maximum effort, at about one-seventh the cost per task.
These are separate evaluations with different workloads and reasoning settings. They support a claim of stronger performance across useful categories, but they do not establish that the two models are interchangeable for every assignment.
What the One-Fifth Price Comparison Means
OpenAI’s API specifications list GPT-6.1 Sol at $2 per million input tokens and $10 per million output tokens. Its Astra pricing is $10 and $50 respectively. On those two standard rates, the new Sol is 80% cheaper.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Sol | $2 | $0.20 | $10 |
| GPT-6.1 Sol | $2 | $0.10 | $10 |
| GPT-6 Astra | $10 | $1 | $50 |
Prices above are per million tokens at standard rates for prompts of up to 272,000 input tokens. Longer requests, processing options, cache writes and tool calls can change the final bill. The comparison concerns API usage, not an 80% reduction in ChatGPT subscription prices.
The unchanged input and output rates also matter when comparing Sol generations. The earlier release had already cut those prices from GPT-5.6 Sol’s promotional levels. GPT-6.1 Sol adds capability at the newer rates and halves cached-input pricing again, from $0.20 to $0.10 per million tokens.
Cheaper Repeated Context Could Help Agents
Repeated context is particularly relevant to an agent that keeps returning to the same instructions, project files or background material. Where requests qualify for caching, the lower read price can reduce the cost of carrying that material through a longer assignment.
That does not make total task cost a fixed multiple of token prices. A model that takes fewer steps, produces less reasoning or needs fewer retries may finish a job more cheaply. Conversely, repeated failed attempts can consume the apparent saving. The useful comparison for a business is the cost of a completed, acceptable result.
AIstify’s coverage of Sol and Luna described the previous release’s emphasis on distributing advanced capabilities across different budgets. The new version continues that strategy by moving the middle tier closer to Astra, without replacing the flagship’s role.
Available in Work and Codex, With Plan Limits
OpenAI’s availability guidance lists Plus, Pro, Business, Enterprise and Edu in the launch rollout. GPT-6.1 Sol is available in ChatGPT Work and Codex, with access depending on the client and workspace settings. It is not yet available in regular Chat, and Free and Go plans are excluded at launch.
Enterprise and Edu administrators must enable the model before members can use it. Standard and Fast modes are available initially, while Ultrafast support is planned for later. Users should therefore distinguish the model becoming available from every speed option reaching every account.
Developers can select gpt-6.1-sol through the API. The model supports a 1.05 million-token context window and up to 128,000 output tokens. Its supported reasoning settings run from low through maximum, and tool calling uses the Responses API. Those details matter for applications migrating from an earlier Sol configuration.
Safety Results Remain Specific to the Tests
The accompanying system card reports fewer unintended outcomes than GPT-6 Sol in deliberately challenging workplace evaluations. In a simulation using 49,650 internal Codex tasks, GPT-6.1 Sol received 28 flags at severity three or above, compared with 42 for GPT-6 Sol and 27 for Astra.
OpenAI defines that severity level as behavior a reasonable user would likely not anticipate and would strongly object to. The company cautions that simulated internal traffic is an additional risk signal, rather than a direct measure of safety in outside deployments.
Progress was not uniform. In a separate test of respecting warnings, unwanted persistence appeared in 23.5% of GPT-6.1 Sol runs, against 17.4% for Astra. That test excluded system-level controls designed to prevent circumvention, so the figures should not be treated as real-world incident rates.
A Faster Upgrade Cycle for OpenAI’s Work Models
The seven-day interval follows the September 22 launch of Sol and Luna. It gives developers another model to evaluate shortly after the previous rollout, particularly where code changes, document handling and repeated tool use dominate workloads.
DevDay also brought OpenAI’s Dots agents, putting ongoing delegated work alongside the model announcements. Together, the releases point to a practical question for customers: how much useful work can an agent complete at an acceptable cost? GPT-6.1 Sol strengthens OpenAI’s case for a less expensive default, while Astra remains the option to compare on the hardest assignments.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
