GPT-6.1 Sol Pricing: Token Rates and Operational Costs

On September 29, 2026, OpenAI introduced GPT-6.1 Sol, an artificial intelligence reasoning model built to deliver complex software engineering, computer interaction, and professional document processing at a fraction of flagship operational expenses. The release establishes standard Application Programming Interface (API) pricing at $2.00 per million input tokens, $0.10 per million cached input tokens, and $10.00 per million output tokens.

Unlike the earlier GPT-6 Sol and the larger GPT-6 Astra systems, GPT-6.1 Sol reduces standard input and output token rates to one-fifth the cost of Astra while closely trailing its benchmark accuracy. On the DeepSWE v1.1 software engineering benchmark, GPT-6.1 Sol matches Astra at roughly 20 percent of the operational cost, while beating the original GPT-6 Sol score by 6.4 percentage points at reduced reasoning effort. Prompt caching now provides a 95 percent discount compared to standard input rates, reducing cached processing costs to half of what OpenAI charged for GPT-6 Sol caching.

For operators managing data pipelines, legal review workflows, or financial extraction pipelines, this release changes how teams budget for document-intensive tasks. The combination of cheap cached inputs and lower per-token rates allows mid-market engineering teams to run automated verification and multi-step tool execution without incurring the prohibitive billing spikes typical of earlier reasoning models.

Official GPT-6.1 Sol Pricing and Token Economics

Standard GPT-6.1 Sol pricing is set at $2.00 per million input tokens and $10.00 per million output tokens for developers querying the API directly. These rates apply to standard text and multimodal requests submitted through the programmatic endpoint under the model identifier gpt-6.1-sol. OpenAI has not published context window limits or maximum output token ceilings for this model version on its announcement page, so engineers should verify technical parameters directly on the official developer platform.

Prompt caching lowers input costs to $0.10 per million tokens for cached prefixes, representing a 95 percent discount against standard input pricing. This discount is 50 percent cheaper than the cached token pricing of the original GPT-6 Sol model introduced a week earlier. For agentic loops and multi-turn workflows where system instructions, schema definitions, and background documents remain consistent across consecutive calls, prompt caching reduces the effective blended cost per query.

OpenAI also announced plans to release a GPT-6.1 Sol Ultrafast variant in Codex. The vendor states that Ultrafast will deliver up to eight times faster token generation speeds compared to standard generation speeds in coding tasks. OpenAI has not published a separate rate card for the Ultrafast variant, meaning developers must verify whether speed enhancements carry a price premium once the feature ships.

• Standard input tokens: $2.00 per million tokens.
• Cached input tokens: $0.10 per million tokens, representing a 95 percent reduction from standard rates.
• Standard output tokens: $10.00 per million tokens.
• API model identifier: gpt-6.1-sol.

Subscription Plan Access across ChatGPT Tiers

ChatGPT Plus, Pro, Business, Enterprise, and Education accounts receive access to GPT-6.1 Sol in ChatGPT Work and Codex starting September 29, 2026. The model is deliberately excluded from the standard consumer ChatGPT Chat interface at launch. OpenAI has not stated whether usage quotas apply to workspace accounts or whether seat prices for ChatGPT Business and Enterprise tiers will increase following this deployment.

Teams using ChatGPT Work can route tasks to GPT-6.1 Sol for collaborative document analysis and project execution without paying per-token API charges. However, high-volume automated routines, background batch pipelines, and custom software integrations require direct API credits through an OpenAI developer organization account. Relying on workspace seats is sufficient for manual team workflows, but production software implementations require budgeting under the standard API rate card.

Developers using Codex receive access to GPT-6.1 Sol for code completion, refactoring, and automated test writing. The upcoming Ultrafast mode will also deploy directly into the Codex environment. Organizations looking to run continuous integration pipelines or automated pull request reviews should plan for API billing, as seat-based subscription interfaces do not support unassisted, headless execution at enterprise scale.

Evaluating GPT-6.1 Sol Pricing against Competing Models

GPT-6.1 Sol pricing operates at exactly 20 percent of the cost of GPT-6 Astra across both input and output dimensions. On technical benchmarks, the price-to-performance gap between the two models narrows substantially. On the GDP.pdf benchmark, which tests question answering across complex Portable Document Format (PDF) files in finance, healthcare, and legal sectors, GPT-6.1 Sol approaches Astra scores while maintaining an 80 percent lower per-task expense.

The model also undercuts competing flagship reasoning systems from rivals like Anthropic. On AutomationBench 1.0.6, an evaluation covering 47 business tools across marketing, finance, and human resources operations, GPT-6.1 Sol scored 2.2 percentage points higher than Claude Opus 5.5 at medium reasoning effort while running at roughly one-third of the cost. OpenAI notes that evaluations for Claude Fable 5.1 exclude fallback costs, which occurred on approximately 40 percent of tested tasks.

Benchmarked task costs in scientific data analysis show a similar margin. On Terminal-Bench Science 0.1, GPT-6.1 Sol completed maximum-effort simulation and theorem-proving tasks at an average cost of $5.47 per task. In comparison, Claude Opus 5.5 averaged $23.21 per task and GPT-6 Astra averaged $23.80 per task. While Astra achieved the top score of 68.1 percent, GPT-6.1 Sol lowered execution costs by over 75 percent while doubling the performance score of the base GPT-6 Sol model.

Worked Monthly Budget for Production Workloads

Calculating monthly expenditure under the GPT-6.1 Sol pricing card requires separating static context from dynamic inputs and reasoning outputs. In a production pipeline processing 10,000 complex document reviews each month, an application might submit 5,000 prompt tokens and generate 1,500 output tokens per request. If 4,000 of those input tokens represent static system instructions, legal taxonomies, and retrieval context, prompt caching absorbs the majority of the input volume.

The monthly math for this 10,000-transaction workload breaks down into three distinct token pools:

Comparing this $200.00 monthly expense against GPT-6 Astra highlights the financial difference. At Astra rates of $10.00 per million input tokens and $50.00 per million output tokens, an equivalent workload without caching discounts would cost approximately $1,250.00 per month. Even with identical prompt caching mechanics, Astra remains approximately five times more expensive to run over high-volume data cycles.

• Cached inputs: 40 million tokens (4,000 tokens x 10,000 requests) at $0.10 per million tokens equals $4.00.
• Uncached inputs: 10 million tokens (1,000 tokens x 10,000 requests) at $2.00 per million tokens equals $20.00.
• Generated outputs: 15 million tokens (1,500 tokens x 10,000 requests) at $10.00 per million tokens equals $150.00.
• Buffer for multi-turn tool loops and retries: roughly $26.00.
• Total estimated monthly spend: $200.00.

Who This Pricing Model Suits and Workload Routing Rules

GPT-6.1 Sol pricing targets production systems that execute multi-step reasoning, computer control, and complex tabular document parsing. If your engineering workload requires reading hundreds of financial prospectuses, navigating software user interfaces via the OSWorld benchmark suite, or running automated code migrations, this model provides the necessary precision without the prohibitive overhead of Astra. Factuality benchmarks show error rates drop to 7.7 percent at low reasoning effort, making the model reliable for regulated data extraction.

This pricing tier is not suitable for basic conversational bots, high-throughput text summarization, or simple classification tasks. Routing simple customer support triage, basic email routing, or single-sentence sentiment tagging to GPT-6.1 Sol wastes capital. Teams handling routine classification should route those requests to smaller, sub-dollar utility models like GPT-4o mini, reserving GPT-6.1 Sol for tasks that require formal logical validation or tool execution.

To maximize cost efficiency, technical leads should implement deterministic routing rules at the gateway layer. Simple requests stay on entry-tier models, while ambiguous inputs, document extractions, and multi-turn agent scripts escalate to GPT-6.1 Sol. In safety evaluations, GPT-6.1 Sol reduced failure rates on broken search disclosures to 2.1 percent compared to 28.7 percent for GPT-6 Luna, making it a reliable tier when security and rule compliance justify the token expense.

Operational Tradeoffs and What Changes the Math

Adopting GPT-6.1 Sol requires balancing raw token savings against task complexity and inference latency. While GPT-6.1 Sol matches or approaches GPT-6 Astra on professional benchmarks like GDP.pdf and DeepSWE v1.1, Astra remains OpenAI's recommended model for the most difficult scientific research tasks where maximum accuracy is mandatory. If an organization runs high-stakes pharmaceutical simulations or theorem validation where a single logical failure compromises the research, paying five times more for Astra remains the correct technical decision.

Several conditions would change this economic assessment. If competing providers reduce prices on high-reasoning models like Claude Opus, or if OpenAI alters the 95 percent caching discount, teams must re-evaluate their pipeline budgets. Similarly, if your application cannot reuse prompt prefixes because input structures vary completely between queries, your effective input cost rises from $0.10 to the full $2.00 per million tokens, eliminating much of the anticipated margin.

Audit your existing token logs to calculate your prompt caching hit rate, verify current figures on OpenAI's official pricing page before finalizing infrastructure budgets, and route your highest-complexity document workloads through a pilot deployment to evaluate actual token consumption against these benchmarks.