GPT-6.1 Sol Economy API

GPT-6.1 Sol is OpenAI's new mid-size flagship model released at DevDay 2026 on September 29, 2026, positioned as "near GPT-6 Astra performance at one-fifth the price." It is engineered for long-horizon, complex end-to-end enterprise workflows: deep multi-step reasoning, advanced software engineering, and native computer-use capabilities (navigating software interfaces via GUI and executing multi-application workflows directly). Powered by symbolic world modeling and dynamic course-correction, it breaks open-ended goals into actionable steps and handles ambiguous edge cases. It sets new benchmarks on ARC-AGI-3 and complex financial, scientific, and coding tasks — matching Astra on DeepSWE v1.1 at roughly one-fifth the cost, and scoring 2.2 points above Anthropic's Opus 5.5 on AutomationBench at about one-third the cost (OpenAI's own evaluation). The Economy tier carries the model's capabilities at even better token prices for volume agent workloads.

The model is offered through the iCreat platform via the OpenAI-compatible Chat Completions API, supporting both streaming and non-streaming modes. Billed by token usage: $0.20 per million input tokens, $1.00 per million output tokens, $0.02 for cache reads, and $0.25 for cache writes.

Model Positioning

GPT-6.1 Sol Economy is the cost-optimal choice for enterprise agent orchestration and autonomous software development. Within OpenAI's GPT-6 family, Astra is the performance summit (officially $10/$50 per million tokens) and Luna the extreme low end ($0.10/$0.50); Sol sits in the middle — approaching Astra's agentic coding and computer-use performance at one-fifth its official price. The Economy tier on the iCreat channel pushes the input price down to $0.20 and output to $1.00, paired with an ultra-low cache-read price, purpose-built for agent workloads that reuse context heavily. Typical scenarios include enterprise agent orchestration, autonomous software development, financial intelligence, and complex legal/business decision support.

Core Capabilities

Long-Horizon Agentic Workflows

Built for multi-step, end-to-end enterprise tasks: open-ended goals are decomposed into actionable steps, complex operations execute across unstructured environments, and dynamic course-correction keeps long chains on track.

Native Computer Use

Navigates software interfaces directly via GUI and executes cross-application workflows — clicking, typing, and operating real software rather than emitting text instructions — the foundation for agents that take over actual software operations.

Agentic Coding and Software Engineering

Matches GPT-6 Astra on the DeepSWE v1.1 software-engineering benchmark at roughly one-fifth the cost, beating GPT-6 Sol's best score by 6.4 percentage points; analyzes projects, locates issues, edits files, and tests code across development phases.

Complex Professional Task Handling

Outscores Opus 5.5 (with fallbacks) on the GDP.pdf professional-document benchmark at less than half the per-task cost; comes within 2.1 points of Astra on OSWorld 2.0 computer use at about one-seventh the per-task cost; averages $5.47 per task on Terminal-Bench Science 0.1 (vs. $23.80 for Astra).

Lower Factual Error Rate

On deliberately difficult prompts at low reasoning effort, the factual-error share fell from 11.4% (GPT-6 Sol) to 7.7%.

Pricing

Billed by token usage; the cache-read price is one-tenth of standard input, a significant advantage for agent workloads that reuse context.

Application Scenarios

• Enterprise agent orchestration: end-to-end automation of multi-step business processes
• Autonomous software development: project-level code analysis, editing, and testing as an agent
• Financial intelligence: analysis of complex financial documents and decision support
• Legal and business decisions: long-document understanding with multi-constraint reasoning
• Cross-application GUI automation: taking over real software operations via computer use

Model Comparison

Same-Series Comparison

Cross-Model Comparison

Same-series data comes from OpenAI's official releases; the AutomationBench cross-model result is OpenAI's own evaluation; this model's pricing follows the iCreat channel price table.

Why Choose GPT-6.1 Sol Economy?

• Near-Astra agentic coding and computer-use capability at a channel input price of $0.20 per million tokens
• Cache reads at $0.02 per million tokens — a major cost drop for context-heavy agent workloads
• Deep multi-step reasoning + symbolic world modeling + dynamic course-correction keeps long-horizon tasks on track
• Native computer use operates software directly via GUI, handling cross-application workflows in one go
• OpenAI-compatible Chat Completions protocol with streaming and non-streaming support — zero-rework integration

Specifications

Architecture

GPT-6.1 Sol was released on September 29, 2026 at DevDay 2026 as an upgrade to GPT-6 Sol (released September 22, 2026). The model advances toward agentic intelligence: symbolic world modeling and dynamic course-correction support long-horizon execution, integrating deep multi-step reasoning, advanced software engineering, and native computer use, setting new benchmarks on ARC-AGI-3 and complex financial, scientific, and coding tasks. Around the release, OpenAI cancelled the planned October GPT-6.1 Astra upgrade after failed safety tests, making Sol the latest representative of the mid-tier; the same event launched the Agents API public beta and always-on Dots agents, with Sol serving as one of the primary models for such agent workloads.

Notes

• model must be the iCreat model code gpt-6.1-sol; other aliases won't resolve
• All other fields follow the OpenAI Chat Completions protocol; existing OpenAI SDKs work directly (point base_url at iCreat)
• Streaming mode (stream: true) pushes deltas via Server-Sent Events — parse accordingly on the client
• thinking and reasoning_effort control thinking behavior; lower reasoning effort reduces cost and latency
• The cache-read price is far below standard input — context-heavy agent scenarios should leverage caching
• The usage field returns token counts for cost accounting
• No two-step async flow: LLMs are synchronous — results return directly, no task_id polling

FAQ

How do I get started with the GPT-6.1 Sol Economy API?​

Register on the iCreat platform and obtain an API Key from the console, then POST to https://api.icreat.ai/v1/chat/completions. All requests are authenticated with the Authorization: Bearer header. Keep your API Key safe and never expose it in client code or public repositories.

How is the cost calculated?​

Billed by token usage: $0.20 per million input tokens, $1.00 per million output tokens, $0.02 per million cache reads, and $0.25 per million cache writes. For example, a call consuming 5,000 input and 2,000 output tokens costs 0.2×5/1000 + 1.0×2/1000 = $0.003; the usage field in the response shows actual consumption.

Does it support streaming?​

Yes. stream: false (default) returns the full JSON in one shot; stream: true pushes incremental content via Server-Sent Events — suited to token-by-token rendering in interactive scenarios.

How do I control costs?​

Three levers: cap generation with max_tokens; lower reasoning effort via reasoning_effort (the factual-error rate at low effort is already down to 7.7%); and hit the cache for context-heavy agent workloads (cache reads at $0.02 — one-tenth of standard input).

How do I choose between this and GPT-6 Astra?​

Astra remains the performance summit for the hardest tasks (68.1% on Terminal-Bench Science). Sol approaches it at roughly one-fifth the price: matched on DeepSWE, within 2.1 points on OSWorld, and at 1/7–1/5 the per-task cost — Sol is the better deal for everyday agent workloads.

What if a call fails?​

Check in order: whether model is gpt-6.1-sol, whether the messages structure is complete (role and content), whether the request body is valid JSON, and whether the API Key is valid. Under the OpenAI-compatible protocol, error formats match OpenAI's official ones, so official error-code guides apply.