What Is GPT-6.1 Sol? Pricing, Specs, and Access

GPT-6.1 Sol Is a 1M-Token Agentic Workhorse — Near-Astra Performance at $2/$10

GPT-6.1 Sol is OpenAI’s proprietary reasoning model for agentic coding, computer use, and professional work, priced at $2 per 1 million input tokens and $10 per 1 million output tokens. Released on September 29, 2026, it supports text and image input, text output, a 1 million-token context window, and up to 131,072 output tokens. It is available through APIs, Codex, and ChatGPT Work, while regular ChatGPT chat access had not arrived as of October 6, 2026.

Key Takeaways

• GPT-6.1 Sol is a multimodal reasoning model designed for Agentic Coding, Computer Use, document analysis, and long-running professional workflows.
• Its standard price is $2 per 1 million input tokens, $10 per 1 million output tokens, and $0.10 per 1 million cached input tokens.
• The model has a 1 million-token input context window and a maximum output of 131,072 tokens.
• OpenAI positions it around Near-Astra Efficiency: capability close to GPT-6 Astra at one-fifth of Astra’s standard token prices.
• Artificial Analysis scored GPT-6.1 Sol Max at 52, one point behind GPT-6 Astra at 53, with a reported cost per task of $0.72 versus $3.26.
• GPT-6.1 Sol is proprietary and closed-weight; no open-source model license is available.

What Is GPT-6.1 Sol?

GPT-6.1 Sol is the cost-focused model in OpenAI’s GPT-6.1 family. It replaced GPT-6 Sol only 7 days after that model’s release and targets workloads where developers need sustained reasoning without paying flagship rates for every step.

OpenAI’s official introduction describes the model as delivering capabilities comparable to GPT-6 Astra. The emphasis is not ordinary chat. Its intended work includes repository investigation, complex refactoring, multi-application agents, browser tasks, and document-heavy professional workflows.

GPT-6.1 Sol is a closed, multimodal reasoning model built to run long, tool-using workflows without Astra-level token prices.

The model also inherits Astra’s safety approach. OpenAI’s system card addendum states that GPT-6.1 Sol uses the same types of training data and the same safeguards stack as GPT-6 Astra. OpenAI classifies it as Critical for cybersecurity capability and High for biological and chemical capability under its Preparedness Framework.

For lineage and positioning across the wider family, see this technical overview of GPT-6.

GPT-6.1 Sol at a Glance

Specification	GPT-6.1 Sol
Developer	OpenAI
Type	Proprietary multimodal reasoning model
Modality	Text and image input; text output
Context window	1 million input tokens
Maximum output	131,072 tokens, commonly described as 128K
Standard input price	$2 per 1 million tokens
Standard output price	$10 per 1 million tokens
Cached input price	$0.10 per 1 million tokens
Release date	September 29, 2026
Availability	API, Codex, ChatGPT Work, and selected integrations
Regular ChatGPT chat	Not available as of October 6, 2026
License	Proprietary; model weights are not published
Parameter count	Not yet confirmed
Architecture	Not yet published

The launch specification combined a 1 million-token context window, 128K output, and pricing of $2/$10 per 1 million input/output tokens.

Source: @haider1

Published launch terms also apply a long-context premium when the prompt exceeds 272,000 tokens. The reported multipliers are 2× for input and 1.5× for output across the entire request. Teams using the full Million-Token Working Set should include that threshold in cost testing.

How GPT-6.1 Sol Works and What Makes It Different

OpenAI has not published GPT-6.1 Sol’s parameter count, model architecture, training compute, or dense-versus-mixture-of-experts design. Its operational behavior is clearer than its internal structure.

Agentic Coding is the primary workload. GPT-6.1 Sol can inspect repositories, plan changes, edit multiple files, run tools, and iterate after tests or reviews. Products expose multiple Reasoning Effort levels, including settings described as Low, Medium, High, XHigh, and Max, although available labels depend on the interface.

Computer Use extends the same reasoning loop to graphical interfaces and browsers. OpenAI’s launch evaluations placed the model about 7 percentage points above GPT-6 Sol on OSWorld 2.0 and approximately 2.1 percentage points behind GPT-6 Astra. Reported task cost was roughly one-seventh of Astra’s.

The Million-Token Working Set lets an agent retain large repositories, documentation collections, or extended task histories. Context capacity does not guarantee perfect retrieval, but it reduces the need to split large projects into many isolated requests.

The Cache Read Discount is central to the economics. Reusing cached context costs $0.10 per 1 million tokens, a 95% reduction from the $2 standard input rate and half GPT-6 Sol’s $0.20 cached-input price. Agent loops repeatedly resend system instructions, code, and documents, so cache-heavy workloads can have much lower blended costs.

Its defining economic feature is the $0.10-per-million-token Cache Read Discount, not a newly disclosed architecture.

Near-Astra Efficiency describes the final distinction. On the DeepSWE v1.1 launch evaluation, GPT-6.1 Sol reached 75.2%, compared with 74.8% for GPT-6 Astra and 68.8% for GPT-6 Sol. At low reasoning effort, the reported share of answers containing a factual error fell from 11.4% with GPT-6 Sol to 7.7% with GPT-6.1 Sol.

These are benchmark results, not a guarantee of identical production quality. Design taste, latency, tool reliability, and instruction following still vary by workload.

What You Can Do With GPT-6.1 Sol

GPT-6.1 Sol is most useful when a task requires repeated reasoning and tool calls rather than a single short answer.

• Repository-scale development: inspect unfamiliar code, trace defects, plan refactors, edit files, and run tests over multiple iterations.
• Browser and desktop workflows: navigate applications, enter data, collect results, and complete multi-step Computer Use tasks.
• Document analysis: process large reports, specifications, contracts, or internal knowledge bases within the Million-Token Working Set.
• Data and professional work: analyze structured information, draft deliverables, and coordinate work across documents and applications.
• Programmatic graphics: community demos include SVG generation, procedural 3D scenes, animations, and small browser games.
• Hybrid orchestration: one test used $0.17 of GPT-6.1 Sol planning to help a local Qwen model complete 3 coding tasks in 43.4 minutes, compared with 114.2 minutes for the local model alone.

Community reports are mixed on creative frontend work. Several testers preferred Claude models for visual design, while Sol performed well on backend coding, browser use, debugging, and analysis. Those observations are useful routing signals, not controlled evaluations.

How GPT-6.1 Sol Compares

At standard API rates, GPT-6.1 Sol costs one-fifth as much as GPT-6 Astra for both input and output tokens.

Model	Input/output per 1M tokens	Cached input	Reported comparison signal
GPT-6.1 Sol	$2 / $10	$0.10	Artificial Analysis score 52; $0.72 per index task
GPT-6 Astra	$10 / $50	$1.00	Artificial Analysis score 53; $3.26 per index task
GPT-6 Sol	$2 / $10	$0.20	Artificial Analysis score 48; replaced after 7 days
Claude Sonnet 5.5	$2 / $10	$0.20	Coding Agent Index score 68 at $14.19 per task versus Sol’s 63 at $1.04

Artificial Analysis’s model results report 58.3 output tokens per second for Sol Max and a 52 Intelligence Index score. Its cost per index task was $0.72, compared with $3.26 for Astra.

Source: @ArtificialAnlys

Arena initially placed GPT-6.1 Sol Max at number 3 in Code Arena: WebDev. It moved to number 4 after Claude Sonnet 5.5 entered the leaderboard. A separate coding-agent evaluation gave Sol 63 points at $1.04 per task and Sonnet 5.5 Max 68 points at $14.19 per task.

The strongest case for GPT-6.1 Sol is near-flagship capability at a mid-tier token price, not universal benchmark leadership.

For more detail on the flagship alternative, see the GPT-6 Astra analysis.

Availability: How to Access GPT-6.1 Sol

GPT-6.1 Sol is released and accessible through multiple hosted routes. Availability inside subscription products can still depend on plan, client version, workspace policy, and regional rollout.

Access route	Status	What it provides
kie.ai API	Available	kie.ai offers GPT 6.1 Sol via API for application and agent integration
OpenAI API	Available	Direct hosted API access using the gpt-6.1-sol model identifier
Codex	Available	Coding-agent workflows with multiple reasoning settings
ChatGPT Work	Available	Complex work across code, applications, and documents
Regular ChatGPT chat	Not yet available as of October 6, 2026	Broader chat rollout was described as coming later
Model weights	Not available	No self-hosted or open-weight release

The launch rollout covered Plus, Pro, Business, Enterprise, and Edu users in supported Work and Codex surfaces. Subscription quotas are separate from per-token API pricing and can change without altering the published API rate.

Initial serving was affected by heavy demand. Users reported about 20 to 25 tokens per second in normal Codex mode and roughly 40 to 45 tokens per second in a faster mode. On October 5, the default subscription speed was increased by about 50%, with reported throughput moving from approximately 30 to 50 output tokens per second.

Ultrafast support for GPT-6.1 Sol was still pending in the latest supplied material. The broader Ultrafast tier is described as supporting up to 8× faster generation and as much as 300 tokens per second, but GPT-6.1 Sol’s launch date, eligibility, and pricing for that tier were not confirmed.

What We Don’t Know Yet

Several production-relevant details remain unpublished or unsettled:

• Architecture and scale: OpenAI has not disclosed parameter count, active parameters, layer design, or training compute.
• Rate limits: Exact requests-per-minute and tokens-per-minute limits by account tier are not confirmed here.
• Ultrafast access: The release date, subscription eligibility, API terms, and sustained throughput remain unconfirmed.
• Regular chat timing: GPT-6.1 Sol had not reached standard ChatGPT conversations by October 6, 2026.
• Stable subscription quotas: Early user reports conflict sharply, ranging from generous usage to half a weekly allowance consumed by one long thread.
• Vision reliability: Text-and-image input is supported, but isolated community reports describe sessions that failed to read images.
• Cross-workload parity: Near-Astra results appear strong in coding and computer use, but they do not establish parity in every domain.

Frequently Asked Questions

What is GPT-6.1 Sol?

GPT-6.1 Sol is OpenAI's proprietary reasoning model for agentic coding, computer use, and complex professional work. It pairs a 1 million-token context window with text-and-image input and text output.

Is GPT-6.1 Sol open source?

GPT-6.1 Sol is not open source. Its weights are not published, and access is provided through hosted products and APIs under proprietary terms.

How much does GPT-6.1 Sol cost?

GPT-6.1 Sol costs $2 per 1 million input tokens, $10 per 1 million output tokens, and $0.10 per 1 million cached input tokens through OpenAI's standard API pricing. Published launch terms apply long-context multipliers above 272,000 prompt tokens, so production users should check current pricing.

What is the GPT-6.1 Sol context window?

GPT-6.1 Sol has a 1 million-token input context window and supports up to 131,072 output tokens. The large window is useful for repositories and document sets, but effective recall still requires workload testing.

How can I access GPT-6.1 Sol?

GPT-6.1 Sol is available through kie.ai's API, OpenAI's API, Codex, and ChatGPT Work. As of October 6, 2026, it had not yet reached regular ChatGPT chat, and plan or workspace settings could affect access.

GPT-6.1 Sol vs GPT-6 Astra: which is better?

GPT-6.1 Sol is the lower-cost choice, while GPT-6 Astra remains the flagship. Sol costs one-fifth of Astra's standard input and output token rates, and Artificial Analysis scored Sol Max at 52 versus Astra at 53, though results vary by workload.

Is GPT-6.1 Sol good for coding?

GPT-6.1 Sol is designed for agentic coding and long-running repository work. OpenAI's DeepSWE launch evaluation reported 75.2%, while independent coding tests show strong cost efficiency but mixed latency and frontend-design results.

What to watch next: Track the regular ChatGPT rollout, the final availability and pricing of GPT-6.1 Sol Ultrafast, and independent evaluations of long-context recall and end-to-end task time. Production teams should also monitor changes to subscription quotas and the 272,000-token pricing threshold.

Building similar coding and computer-use agents? On kie.ai you can try GPT 6.1 Sol, GPT-6 Astra, and Claude Sonnet 5.5.