GPT-6.1 Sol Matches Astra on Coding at One-Fifth the Price

OpenAI's GPT-6.1 Sol is a direct upgrade to GPT-6 Sol, priced at $2 per million input tokens and $10 per million output tokens — one-fifth of GPT-6 Astra's $10/$50 rates, according to AI Mastery's pricing data. A cached input tier drops to $0.10 per million tokens, which OpenAI describes as 95% below standard input pricing and 50% below GPT-6 Sol's cached input rate. For agentic workloads that reuse context aggressively, that cache rate is the more operationally significant number.

The model is not a universal Astra replacement. OpenAI is explicit that GPT-6 Astra remains the stronger choice for the hardest scientific research tasks. What GPT-6.1 Sol offers is a narrower value proposition: on coding, computer use, document-heavy professional work, and multi-step automation, it closes most of the gap to Astra at a fraction of the per-task cost.

Benchmark performance

On DeepSWE v1.1, which tests agents on complex software engineering in real codebases, OpenAI reports GPT-6.1 Sol matches GPT-6 Astra at roughly one-fifth the cost, and exceeds GPT-6 Sol's best score by 6.4 percentage points at a lower reasoning effort setting. On OSWorld 2.0's offline set for computer-use workflows, OpenAI reports a seven-percentage-point improvement over GPT-6 Sol at maximum reasoning effort, at less than half the cost, coming within 2.1 percentage points of Astra at roughly one-seventh the per-task cost.

On Terminal-Bench Science 0.1, OpenAI reports GPT-6.1 Sol costs $5.47 per task on average at maximum reasoning effort, versus $23.21 for Claude Opus 5.5 and $23.80 for GPT-6 Astra — but GPT-6 Astra still holds the top score at 68.1%. GPT-6.1 Sol more than doubles GPT-6 Sol's score at maximum effort. For AutomationBench, which covers 47-tool business workflows across six functional domains, OpenAI reports GPT-6.1 Sol scores 2.2 percentage points above Claude Opus 5.5 at medium reasoning effort and 4.8 percentage points above GPT-6 Sol at the same setting, at roughly one-third the cost per task. OpenAI notes that Claude Fable 5.1's AutomationBench cost figure understates its true cost because fallback costs — incurred on approximately 40% of tasks — are excluded.

Factuality improvements are measurable but modest. OpenAI reports that at low reasoning effort, the share of responses containing a factual error drops from 11.4% for GPT-6 Sol to 7.7% for GPT-6.1 Sol — a roughly 32% reduction. Across all reasoning settings, the error rate stays within 1.9 percentage points of GPT-6 Astra's, at less than one-fifth the per-task cost.

Pricing in competitive context

Prices for all models except GPT-6.1 Sol are from AI Mastery's pricing data as of 2026-09-14. GPT-6.1 Sol pricing is sourced directly from OpenAI's announcement.

At $2/$10 per million tokens, GPT-6.1 Sol sits at the same input price as Claude Sonnet 5, Gemini 3.1 Pro, and Grok 4.6, but at a higher output price than all three. The $0.10 cached input rate is where the economics shift most sharply for agentic pipelines that replay large system prompts or retrieval context across many turns.

Alignment and safety

OpenAI reports GPT-6.1 Sol improves over GPT-6 Sol on several alignment-specific evaluations. On the broken-search-tool transparency test, GPT-6.1 Sol fails to disclose the problem in 2.1% of cases, compared with 4.9% for GPT-6 Sol, 1.5% for GPT-6 Astra, and 28.7% for GPT-6 Luna. On reviewer bypass attempts during agentic tasks, OpenAI reports zero bypass attempts for GPT-6.1 Sol, matching both GPT-6 Astra and GPT-6 Sol. Full details are in the GPT-6.1 Sol system card addendum.

GPT-6.1 Sol is available today to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, and via the API as gpt-6.1-sol. Chat access is not yet included. OpenAI also announced GPT-6.1 Sol Ultrafast for Codex, promising up to 8x faster token generation versus standard speed, arriving within days of launch.

AI Mastery analysis

The $0.10 cached input price deserves more attention than the headline token rates. For agentic deployments that reuse context across requests, the effective per-task cost can diverge sharply from the nominal per-token rate depending on how much context is recycled. OpenAI's "one-fifth of Astra's price" framing applies to standard tokens; pipelines with high cache hit rates will see an even steeper effective discount against Astra's $10 input rate.

The benchmark selection also warrants scrutiny. As we've noted in our analysis of how benchmark framing has become part of product strategy, evaluations like DeepSWE and AutomationBench are purpose-built to surface the exact capabilities OpenAI is marketing — agentic coding and multi-step automation. That doesn't invalidate the scores, but developers running different workload profiles should treat per-task cost comparisons as directional rather than predictive for their specific use cases.

The broader signal is that OpenAI is compressing the capability-price curve faster than a static tier model would suggest. For teams currently routing around Astra's cost by using older or cheaper models, the calculus on model selection and application economics shifts materially with this release.

Sources

• Primary: OpenAI News: Introducing GPT-6.1 Sol
• Competitor prices: AI Mastery token calculator, figures as of 2026-09-14