OpenAI Launches GPT-6.1 Sol at $2 Input and $10 Output per Million Tokens

GPT-6.1 Sol launched on September 29

OpenAI published GPT-6.1 Sol on September 29, 2026 at 10:00 UTC. Its official feed describes the release as near-Astra intelligence for coding, computer use and professional work at one-fifth of Astra’s standard API input and output token prices. The live API identifier is gpt-6.1-sol.

OpenAI’s current model guide places Sol between GPT-6 Astra, its highest-intelligence option, and GPT-6 Luna, its fastest and lowest-cost family member. “Near-Astra performance” is vendor positioning, not an independent result. The accessible first-party documentation does not establish how close the models are on a reader’s workload, and AccessAllGPT did not run either model.

The headline rate is $2 in and $10 out

For short-context Standard requests, OpenAI lists $2 per million input tokens, $0.10 per million cached-input tokens, $2.50 per million cache-write tokens and $10 per million output tokens. GPT-6 Astra is listed at $10 input and $50 output, which supports OpenAI’s one-fifth token-rate comparison. It does not prove one-fifth the cost for a completed task because models may use different reasoning, output, retry and tool-call budgets.

Fast mode costs twice Standard. Batch and Flex are 50% below Standard, and regional processing adds 10% where available. Cache writes cost 1.25 times uncached input while cache reads cost 5% of uncached input. Record the actual processing mode, region and cache behavior in any cost comparison rather than quoting the base rate alone.

Long prompts move the entire request into a higher price band

GPT-6.1 Sol has a 1,050,000-token context window, with up to 922,000 input tokens and 128,000 output tokens. For a request containing more than 272,000 input tokens, OpenAI says input and cache rates double and output rises by 50% for the full request. The resulting Standard rates are $4 input, $0.20 cached input, $5 cache writes and $15 output per million tokens.

That threshold creates a discontinuity, not a marginal surcharge only on tokens above 272K. Retrieval and agent systems should measure the complete assembled request after instructions, history, documents and tool results are included. A large context ceiling is capacity, not evidence of reliable fact use across every position or of economical full-corpus prompting.

Tool use belongs on the Responses API

OpenAI documents Responses and Chat Completions support, but tool calling is available only through Responses. The supported Responses tools include web search, file search, image generation, Code Interpreter, hosted shell, apply patch, skills, computer use, MCP and tool search. Chat Completions remains available for requests without tools.

Reasoning effort supports low, medium, high, xhigh and max; medium is the default, while none and minimal are unsupported. An evaluation that omits effort is therefore evaluating medium—not a neutral no-reasoning baseline. Pin effort explicitly and cap output, tool calls, elapsed time and spend when comparing quality or cost.

Residency and endpoint limits narrow the deployment choices

The model reference says GPT-6.1 Sol supports US and EU data residency, but Fast mode is unavailable with EU residency. Regional processing carries the documented 10% premium where offered. Eligibility and contractual controls remain account-specific; the presence of an EU option on a model page is not by itself evidence that a particular organization’s retention, subprocessors or residency requirements are satisfied.

The model accepts text and image input and produces text. It does not support audio or video input, fine-tuning, Realtime, Assistants or predicted outputs. Batch is supported. These interface boundaries matter during migration: a system using tool-enabled Chat Completions, a fine-tuned model or Fast mode with EU residency cannot switch by changing only the model string.

Compare complete tasks before replacing Astra or GPT-6 Sol

Build a frozen set of representative coding, computer-use and professional tasks. Run the current model, GPT-6.1 Sol and Astra with explicit effort, the same tools and equivalent stop conditions. Score accepted outcomes, not benchmark impressions. Capture first-token and completion latency, input by price band, cache writes and reads, reasoning and visible output, tool calls, retries, refusals and cost per accepted task.

Pilot when the lower token rate could materially change economics and the Responses API fits the tool contract. Wait when account access, residency eligibility, long-context behavior or migration requirements are unverified. Reject unattended consequential actions without deterministic authorization, least privilege, approval and rollback. The immediate decision is whether GPT-6.1 Sol earns a bounded comparison—not whether “near-Astra” settles the production choice.

Disclosures

AccessAllGPT did not receive OpenAI credentials, credits, early access, a briefing, benchmark data, review or compensation for this article. OpenAI did not sponsor, review or endorse it. AccessAllGPT Research is operated by NeuralArc, is independent, and is not affiliated with OpenAI. Publication-wide relationships are listed on the disclosures page.