GPT-6.1 Sol Pricing: Near-Astra at Sol Rates

OpenAI launched GPT-6.1 Sol on September 29, 2026 as its new balance of frontier capability and cost. The API model ID is gpt-6.1-sol; the live table, chart, and calculator above use the verified rate card rather than prices copied into this article.

The practical headline is narrower than “a cheaper Astra.” GPT-6.1 Sol keeps GPT-6 Sol’s standard fresh-input, cache-write, and output rates while cutting cache reads in half. OpenAI reports near-Astra results on selected agentic evaluations, but Astra still leads the hardest scientific benchmark in the launch material.

What changed

GPT-6.1 Sol accepts text and images, produces text, and has a 1.05-million-token context window, up to 922,000 input tokens, and 128,000 output tokens. Its knowledge cutoff is April 30, 2026.

Reasoning supports low, medium, high, xhigh, and max; medium is the default. Responses supports tool calling, while Chat Completions is supported without tool calling. The model supports structured outputs, prompt caching, web and file search, code execution, computer use, MCP, skills, and image generation as a tool. It does not support fine-tuning, Realtime, audio, or video endpoints.

GPT-6.1 Sol pricing impact

The live comparison shows every current token category. Standard prices remain unchanged from GPT-6 Sol except cached input, which is 50% lower. That makes the upgrade most valuable for agents that repeatedly send stable instructions, tool schemas, repository maps, or document prefixes.

Requests above 272,000 input tokens reprice the full request: input and cache categories double, while output rises by half. Batch and Flex are 50% below Standard; Fast mode is 2× Standard. Eligible regional processing adds 10%, and Fast mode is unavailable with EU data residency.

OpenAI says GPT-6.1 Sol Ultrafast is coming to Codex with up to 8× faster token generation. No Ultrafast price for this model is published yet, so do not budget it at Standard or Fast rates.

Use the token calculator for your traffic mix and the complete OpenAI pricing page for current model rows.

What the evidence does—and does not—show

OpenAI says GPT-6.1 Sol matched Astra on DeepSWE at roughly one-fifth of the tested cost. It reports a 6.4-point improvement over GPT-6 Sol, a 7-point Sol gain on OSWorld, and a factual-error rate that fell from 11.4% to 7.7% at low effort. On AutomationBench, OpenAI reports a 2.2-point lead over Opus 5.5 at medium effort. Astra retained the highest reported Terminal-Bench Science score.

Our launch-day openai/gpt-6.1-sol route scored 49/49 with zero errors on the maintained deterministic Labs suite. The current full-roster run used 3,426 prompt tokens and 1,004 completion tokens, costing $0.016892 total, or $0.00034473 per correct answer. Exact, numeric, regex, containment, and structured-field graders required no LLM judge.

That run verifies route availability and narrow deterministic reliability. It does not reproduce OpenAI’s coding, computer-use, science, or professional-work tests. OpenAI’s results remain vendor-run and configuration-specific, not proof that every workload becomes five times cheaper.

What DevDay 2026 adds

OpenAI’s DevDay recap says the event included more than 20 announcements across ChatGPT, Codex, models, and new work patterns. It highlighted agents that can own ongoing responsibilities and a more open ChatGPT surface where developers can launch native experiences to OpenAI’s reported 1.2 billion weekly users.

For API buyers, the verified near-term items are GPT-6.1 Sol and the coming Ultrafast Codex route. The recap does not publish a separate GPT-6.1 Sol subscription price, Ultrafast rate, or API surcharge. ChatGPT Work and Codex access is rolling out to Plus, Pro, Business, Enterprise, and Edu users; Chat access is not yet available.

Who benefits—and who should wait

Likely winners: coding agents, document workflows, computer-use systems, and repeated-context agents that can exploit cheaper cache reads. Teams paying Astra rates for tasks that Sol nearly matches now have an obvious evaluation candidate.

Keep Luna: routine extraction, classification, support routing, and high-volume tasks with easy verification.

Keep Astra: difficult scientific, research, or end-to-end tasks where your acceptance rate falls on Sol.

Wait: regulated teams that require Fast mode with EU residency, Chat users, and anyone budgeting Ultrafast before its price appears.

What developers should do now

1. Add gpt-6.1-sol as a new route; do not silently repoint gpt-6-sol.
2. Replay the same prompts, tools, effort, stop rules, and context lengths against Sol and Astra.
3. Track accepted results, cache hit rate, retries, latency, tool fees, and reviewer time.
4. Test below and above the long-context threshold separately.
5. Start balanced agent traffic on GPT-6.1 Sol, keep Luna below it, and escalate measured failures to Astra.

Compare the live Anthropic pricing page and, for an OpenAI-compatible open-model control, benchmark Novita routes on the same accepted-result rubric.

Affiliate disclosure: AI Pricing Guru may earn a commission from the sponsored Novita link at no extra cost to you. It does not affect this analysis.

Bottom line

GPT-6.1 Sol is now OpenAI’s balanced model for complex work: the upgrade preserves Sol’s standard rate card, improves cache economics, and narrows the reported capability gap with Astra. It is a migration candidate, not an automatic global replacement.

OpenAI’s GPT-6.1 Sol announcement, model documentation, API pricing, system-card addendum, and DevDay 2026 recap were checked September 29, 2026. For the previous generation, see GPT-6 Sol and Luna launch pricing.