OpenAI's GPT-6.1 Sol Delivers Astra Performance at a Fifth of the Cost

30-Second Summary

OpenAI released GPT-6.1 Sol on September 29, 2026, at DevDay 2026. The model matches its flagship GPT-6 Astra for agentic coding, computer use, and professional work at one-fifth of Astra's token price. Input costs $2 per million tokens, output costs $10, and cached inputs cost $0.10, which is 95 percent below the standard rate. The model is live now through the API and ChatGPT Work.

At a Glance

• Topic: Model Releases
• Company: OpenAI
• Date: September 29, 2026
• Announcement: GPT-6.1 Sol released at DevDay 2026 with near-Astra performance at 80 percent lower input and output cost than GPT-6 Astra
• What Changed: Enterprise and developer access to Astra-class reasoning is now 5x cheaper, with a 1.05 million token context window and a 95 percent discount on cached inputs
• Why It Matters: The cost barrier for deploying high-capability agents across real business workflows has dropped sharply in a single release
• Who Should Care: Any operator, developer, or business leader running or planning agentic workflows, automation pipelines, or code generation at scale

Key Facts

• Pricing: $2.00 per million input tokens, $10.00 per million output tokens
• Cached input: $0.10 per million tokens (95 percent below standard input, 50 percent below GPT-6 Sol's cached rate)
• Context window: 1,050,000 tokens
• Max output: 128,000 tokens
• Knowledge cutoff: April 30, 2026
• Availability: API (gpt-6.1-sol), ChatGPT Work and Codex, accessible to Plus, Pro, Business, Enterprise, and Edu subscribers
• Not available: ChatGPT Chat interface at launch
• Benchmark: Matches GPT-6 Astra on DeepSWE v1.1 (complex software engineering tasks), and exceeds GPT-6 Sol's best score by 6.4 percentage points at lower reasoning effort and cost
• Ultrafast variant: GPT-6.1 Sol Ultrafast, offering up to 8x faster token generation in Codex, expected within days of launch

What Happened

OpenAI held its annual DevDay developer conference on September 29, 2026, seven days after releasing GPT-6 Sol. The headline release was GPT-6.1 Sol, a model the company describes as delivering near-Astra intelligence for agentic coding, computer use, and professional work at one-fifth of GPT-6 Astra's token costs.

The pricing is the central claim. GPT-6 Astra, OpenAI's most capable model, costs $10 per million input tokens and $50 per million output tokens. GPT-6.1 Sol costs $2 and $10 respectively, placing Astra-class performance within reach of workflows that were economically unfeasible at Astra's pricing. Cached inputs drop further to $0.10 per million, which is 50 percent below the cached rate for GPT-6 Sol.

On DeepSWE v1.1, an industry benchmark measuring performance on complex software engineering tasks in real codebases rather than synthetic tests, GPT-6.1 Sol matches GPT-6 Astra's score and exceeds GPT-6 Sol's best result by 6.4 percentage points at lower reasoning effort. The model supports a 1.05 million token context window and can return up to 128,000 tokens per call. An Ultrafast variant with up to 8x faster generation in Codex is expected within days of launch.

Access opened immediately on September 29 via the API and through ChatGPT Work and Codex. It is not available in the standard ChatGPT Chat interface. All Plus, Pro, Business, Enterprise, and Edu accounts have access.

Why It Matters

Agentic work becomes affordable for small teams. The 5x cost reduction on Astra-class performance means that multi-step agent tasks such as autonomous code review, iterative document analysis, and research synthesis move from premium experiments to routine deployments for businesses with modest AI budgets.

Cached input pricing resets architecture decisions. At $0.10 per million cached tokens, any architecture that repeatedly injects a large system prompt or knowledge base becomes dramatically cheaper. Retrieval-augmented applications, persistent agent contexts, and compliance check pipelines all benefit without code changes.

The 1.05 million token context window removes chunking overhead. For operators processing long documents, including contracts, board papers, regulatory submissions, and large codebases, a context window of this size means entire document sets can be processed in a single call. The operational complexity of splitting, tracking, and reassembling chunks disappears.

GPT-6.1 Sol introduces a new tier in the OpenAI model family. OpenAI now has two distinct price and performance points at the high end: Astra for tasks that require maximum capability, and Sol at 80 percent lower cost for work where Astra-level performance is sufficient. Operators can route tasks between tiers rather than paying Astra prices for everything.

The benchmark result shifts competitive positioning. A model that matches Astra on DeepSWE v1.1 at one-fifth the cost creates pressure across the market. Competing providers running Astra-class equivalents at higher prices will face a measurable cost disadvantage in developer and enterprise sales.

The David and Goliath View

OpenAI's DevDay 2026 delivered the most commercially significant pricing shift in agentic AI this year. The gap between what top-tier models can do and what operators can afford to run them on has been the practical constraint on AI deployment for most businesses. GPT-6.1 Sol does not close that gap entirely, but it narrows it substantially and immediately.

The implication for businesses considering agentic automation is direct. Workflows that were previously modelled at Astra pricing and found too expensive should be revisited this week. The numbers changed. A research loop that cost $500 per run at Astra rates costs $100 at Sol rates. That is the difference between a quarterly experiment and a weekly production process.

The cached input pricing deserves particular attention. At $0.10 per million tokens, building systems with persistent context, large instruction sets, or knowledge bases loaded at every call stops being a budget concern and becomes an architecture choice. Operators who have been keeping their system prompts lean for cost reasons now have room to build richer, more capable agents without the token maths fighting against them.

Where This Fits in the AI Stack

GPT-6.1 Sol sits at the intersection of the model layer and the automation layer. It does not change what is possible, but it changes the cost at which routine agentic tasks run. For operators building on the AI Growth Engine or Employee Amplification Systems, this release affects the per-task economics of every workflow that involves multi-step reasoning, code generation, or large document processing. The practical upgrade path is straightforward: test existing Astra-powered pipelines on gpt-6.1-sol, verify output quality holds, and reprice accordingly.

Questions Operators Are Asking

Can I switch my existing GPT-6 Astra API calls directly to GPT-6.1 Sol? Yes. The model is available via the API as gpt-6.1-sol and accepts the same input formats. Output quality for agentic coding and professional work tasks is reported to match Astra on benchmarks, but you should run your own eval before moving production traffic. The API parameter change itself is a single line.

Does GPT-6.1 Sol handle the same context length as Astra? GPT-6.1 Sol supports 1,050,000 tokens of context and 128,000 tokens of output. Whether this matches Astra exactly depends on the specific Astra configuration your account uses. For most enterprise use cases involving document analysis or extended agentic sessions, GPT-6.1 Sol's context window is sufficient.

Is the 95 percent cached input discount applied automatically? Caching works automatically for repeated prompt prefixes above a minimum length. OpenAI's API documentation covers the exact mechanics. If your system prompt or knowledge context exceeds a few thousand tokens and is consistent across calls, you are likely already eligible for cached pricing without any code change.

What tasks is GPT-6.1 Sol not suited for? OpenAI positioned the model specifically for agentic coding, computer use, and professional work. Tasks requiring multimodal video reasoning or the highest possible benchmark scores on non-coding domains may still benefit from Astra. The DeepSWE result covers software engineering specifically, not the full range of enterprise tasks.

When is the Ultrafast variant available? OpenAI stated at DevDay that GPT-6.1 Sol Ultrafast would follow within days of the September 29 launch. As of October 5, 2026, no separate release announcement has been made. Monitor OpenAI's changelog or the gpt-6.1-sol-ultrafast model ID in the API for availability.

Citable Summary

OpenAI released GPT-6.1 Sol on September 29, 2026, at DevDay 2026. The model delivers near-Astra performance for agentic coding, computer use, and professional work at $2 per million input tokens and $10 per million output tokens, compared to GPT-6 Astra's $10 and $50. Cached inputs cost $0.10 per million tokens. The model features a 1,050,000 token context window, matches GPT-6 Astra on the DeepSWE v1.1 software engineering benchmark, and is available now through the OpenAI API and ChatGPT Work. An Ultrafast variant with up to 8x faster generation in Codex is expected within days.