One point. That’s the entire gap between OpenAI’s new GPT-6.1 Sol and its flagship GPT-6 Astra in the Artificial Analysis Intelligence Index — at less than a quarter of the Cost per Task. OpenAI launched the model at DevDay on September 29, 2026, and the pitch is simple: near-flagship intelligence for a fifth of the price.
The company didn’t stop at performance claims. On X, OpenAI called GPT-6.1 Sol “the most cost-efficient model for its performance available today.” A bold statement, especially given what happened to its predecessor just a week earlier.
TL;DR: OpenAI launched GPT-6.1 Sol at DevDay on September 29, 2026, an upgrade to GPT-6 Sol that nearly matches GPT-6 Astra in agentic coding, computer use, and professional work. Artificial Analysis measured it just 1 point below Astra in the Intelligence Index at less than one quarter of the Cost per Task — roughly a fifth of Astra’s token price. In the API it costs $2 per million input tokens and $10 per million output tokens, with a new Ultrafast tier running at 300 tokens per second.
What Is GPT-6.1 Sol and Why Did OpenAI Release It?
GPT-6.1 Sol is an upgraded version of GPT-6 Sol, announced at OpenAI’s DevDay 2026 event on September 29, 2026. According to OpenAI, the model focuses on enhancing agentic coding, computer use, and professional work capabilities. The goal is straightforward: deliver performance close to the GPT-6 Astra flagship while cutting the price dramatically.
Why does this tier exist at all? Coverage of the launch points to a clear market logic. The most expensive model from OpenAI is no longer the only sensible choice for demanding users — one Polish outlet, Telepolis.pl, described the release as delivering a “golden middle” between price and capability. VentureBeat reporting frames the target audience even more precisely: applications where waiting carries its own cost, including interactive coding, customer support, financial analysis, incident response, and other human-in-the-loop systems.
The model is already broadly available. Amazon announced that GPT-6.1 Sol is generally available on Amazon Bedrock, bringing stronger reasoning to coding, computer use, and professional workloads that run frequently — exactly the kind of high-volume use cases where token pricing matters most.
For paying ChatGPT users, the model arrived on a Tuesday alongside the API release, with early coverage noting strong results in coding jobs, dense document reading, and long-context work. OpenAI also paired the launch with new Codex and ChatGPT tools, though details on those ship separately from the model itself.
How Close Does GPT-6.1 Sol Get to GPT-6 Astra in Benchmarks?
Artificial Analysis put a precise number on the gap: GPT-6.1 Sol scores just 1 point below GPT-6 Astra in the Intelligence Index, while costing less than one quarter of the Cost per Task. That framing — near-Astra performance at a fraction of the price — is the core of the entire launch.
One caveat matters here. Much of the “comes close to Astra” messaging relies on OpenAI’s own benchmarks, as The Decoder points out in its coverage. Vendor-run evaluations deserve healthy skepticism, and independent measurement from Artificial Analysis is what gives the claim real weight. Fortunately, that independent check exists and largely agrees with the company’s story.
The benchmark picture is not a single aggregate number alone. Sources describe gains across several specific dimensions: major improvements in coding, computer use, scientific research, and factual accuracy, according to Digital Trends. On the agentic side, The New Stack reports the model nearly matches GPT-6 Astra specifically on agentic coding and computer use benchmarks — the two areas most relevant to developers building autonomous workflows.
How should developers read a one-point difference? In practical terms, it suggests the flagship’s remaining advantage is thin. Whether that thin margin justifies a fivefold price difference is exactly the calculation OpenAI wants its customers to make — and wants them to answer in Sol’s favor.
Why Did GPT-6 Sol Last Only Seven Days?
GPT-6.1 Sol replaces GPT-6 Sol after just seven days. That’s not a typo or a rumor — it’s the headline of Artificial Analysis’s own coverage of the launch. A model surviving a single week in the lineup before being superseded is a remarkably short lifecycle, even by the accelerated pace of current AI releases.
The sequence of events matters for understanding the situation. OpenAI shipped GPT-6 Sol first, and a week later — at DevDay on September 29, 2026 — it introduced GPT-6.1 Sol as an upgrade. The New Stack confirms the timing directly: GPT-6.1 Sol lands a week after GPT-6 Sol. OpenAI describes the new model as delivering significant improvements over GPT-6 Sol across complex professional tasks, including code writing and debugging, document understanding, and executing multi-step tasks.
What does a seven-day replacement cycle mean for developers? Three things stand out:
- Pricing commitments made against GPT-6 Sol may need re-evaluation, since the upgraded model changes the cost-performance equation entirely
- Benchmark comparisons should track the .1 model going forward, as the original Sol’s numbers no longer represent the current offering
- Model naming and versioning in the API need to account for rapid supersessions when building production dependencies
Neither OpenAI nor the coverage indicates that GPT-6 Sol remains available or has been deprecated — sources simply note the replacement. The planned flagship, GPT-6.1 Astra, meanwhile, is staying under wraps for now, according to The Decoder.
What Do GPT-6.1 Sol’s Improvements Cover in Coding and Computer Use?
The upgrade concentrates on the agentic core: coding, computer use, and professional work. OpenAI says GPT-6.1 Sol delivers significant improvements over GPT-6 Sol specifically in code writing and debugging, document understanding, and executing multi-step tasks — the three pillars of real-world agent behavior.
Each of these areas maps to concrete developer workflows. Agentic coding means the model can plan and execute development tasks rather than just autocomplete snippets. Computer use means it can operate software interfaces directly. Document understanding covers dense document reading, an area where early ChatGPT user coverage highlighted strong performance. Multi-step execution ties everything together, since agentic systems live or die on whether a model can sustain coherent work across chained actions.
Beyond the agent-focused headline features, the improvement list is broader. Digital Trends reports major gains in scientific research and factual accuracy on top of coding and computer use. Amazon’s Bedrock announcement emphasizes stronger reasoning for frequently-run professional workloads — a signal that OpenAI optimized this model for the repeated, high-volume tasks that dominate enterprise API bills rather than one-off showcase queries.
VentureBeat’s analysis of the target use cases reinforces this: interactive coding, customer support, financial analysis, and incident response are all domains where multi-step professional work and low latency matter together. The improvement profile and the intended market line up cleanly.
How Much Does GPT-6.1 Sol Cost in the API?
GPT-6.1 Sol is available in the API at $2 per million input tokens and $10 per million output tokens, according to Unite AI’s coverage of the DevDay 2026 announcement. That pricing underpins the “one-fifth of the price” claim relative to GPT-6 Astra’s token costs.
The pricing story includes a second layer: speed. VentureBeat reports a new Ultrafast tier that clocks at 300 tokens per second. For latency-sensitive applications — live coding assistants, support chats, incident response consoles — raw throughput matters as much as price, and a dedicated fast tier targets exactly those deployments.
Artificial Analysis measures cost differently from the sticker price, using its Cost per Task metric. By that measure, GPT-6.1 Sol comes in at less than one quarter of GPT-6 Astra’s cost — consistent with OpenAI’s marketing claim of roughly one-fifth the token price. The slight difference between “less than a quarter” and “a fifth” reflects the gap between theoretical token pricing and measured real-world task completion cost, which includes how many tokens a model actually consumes to finish a task.
For teams weighing the flagship against this tier, the economic argument is now explicit in OpenAI’s own words: near-Astra intelligence for a fifth of the price, with the company calling it the most cost-efficient model for its performance available today. The remaining question — where the flagship still earns its premium — is one OpenAI will answer when GPT-6.1 Astra eventually emerges from under wraps.
What Is the New Ultrafast Tier and Who Is It For?
The Ultrafast tier is a new serving option for GPT-6.1 Sol that clocks at 300 tokens per second, according to VentureBeat’s coverage of the launch. That throughput figure is the headline number for the tier, and it signals OpenAI’s intent to compete on latency, not just price. For teams billing by the interaction rather than the token, speed is a feature.
Coverage indicates the obvious targets are applications where waiting carries its own cost: interactive coding, customer support, financial analysis, incident response, and other human-in-the-loop systems. In each of these scenarios, a slow model response breaks the user’s flow or delays a decision that has real financial consequences.
Who should actually consider Ultrafast? A few groups stand out:
- Developers using AI pair-programming tools, where every pause between prompts compounds into lost focus
- Support teams running live chat assistants who cannot leave customers waiting through long generations
- Analysts working through dense documents who iterate quickly and expect rapid turnarounds
- Incident response teams that need reasoning-quality output under time pressure
- Platforms building human-in-the-loop review workflows where the human is idle until the model responds
The combination matters. Near-Astra intelligence alone would be interesting; near-Astra intelligence at 300 tokens per second turns a batch-oriented model into a conversational one. Sources suggest this tier is a deliberate play for workloads that flagship models handle well but at price and speed points that make everyday use impractical.
Where Can Developers Run GPT-6.1 Sol Beyond OpenAI’s Platform?
GPT-6.1 Sol is now generally available on Amazon Bedrock, according to an AWS Machine Learning blog post published around the launch. The post describes the model as bringing stronger reasoning to coding, computer use, and professional workloads that run frequently — exactly the high-volume categories where price differences compound into large monthly bills.
This matters for enterprises that cannot or will not route traffic directly through OpenAI’s API. Bedrock availability means teams already committed to AWS can call GPT-6.1 Sol through the same infrastructure, billing, and governance pipelines they use for the rest of their machine learning stack. Procurement and security reviews get simpler. That alone can shorten adoption timelines.
For architects evaluating placement, the practical picture looks like this:
| Deployment Path | What Sources Confirm |
|---|---|
| OpenAI API | Available at launch, priced at $2 per million input tokens and $10 per million output tokens (unite.ai) |
| Amazon Bedrock | Generally available, positioned for frequent coding, computer use, and professional workloads (AWS blog) |
| ChatGPT | Available to paying ChatGPT users (TechEBlog) |
| Codex and ChatGPT tooling | New tools introduced alongside the model at DevDay (unite.ai) |
One caveat: source coverage of Bedrock availability does not detail region-by-region rollout or enterprise pricing differences, so teams should verify terms directly on the AWS blog and Bedrock documentation before committing production traffic.
What Happened to GPT-6.1 Astra at DevDay?
GPT-6.1 Astra did not launch at DevDay. The Decoder reports that OpenAI’s planned flagship is staying under wraps for now, even as the company shipped its cheaper sibling with near-flagship performance claims. OpenAI’s own announcement framing — “near-Astra intelligence for a fifth of the price” — quietly confirmed that Astra remains the reference point rather than the product of the day.
The decision is telling. Instead of leading with the most capable model, OpenAI led with the most economical one. The company’s X post called GPT-6.1 Sol “the most cost-efficient model for its performance available today,” a claim aimed squarely at buyers who benchmark on dollars per unit of intelligence rather than raw leaderboard position.
There are at least two plausible readings, and sources leave room for both:
- Astra is being held for a separate, higher-profile launch, keeping DevDay focused on developer economics rather than frontier claims
- The Sol release gives OpenAI a revenue-generating middle tier while Astra undergoes whatever additional preparation the safety lead and remaining teams require — The Decoder notes a safety lead in its coverage, though the snippet does not detail their statements
What is verifiable is the sequencing. GPT-6.1 Sol shipped publicly on September 29, 2026, at DevDay, with API availability, Bedrock availability, paying ChatGPT access, and new Codex tooling. Astra’s GPT-6.1 generation, by contrast, exists in coverage only as a benchmark target and a promise. Until OpenAI reveals it, every “near-Astra” claim rests on the company’s own benchmarks.
How Does GPT-6.1 Sol Change the Cost-Efficiency Race?
GPT-6.1 Sol scores 1 point below GPT-6 Astra in the Intelligence Index while costing less than one quarter of Astra’s Cost per Task, according to Artificial Analysis. OpenAI’s own framing is more aggressive: one fifth of the token price. Either way, the gap between flagship capability and mid-tier pricing has narrowed to a rounding error on quality and a chasm on cost.
The speed of this turnover is just as notable as the numbers. Artificial Analysis points out that GPT-6.1 Sol replaced GPT-6 Sol after just 7 days — the previous Sol generation barely had a week in the market before its successor arrived. Hacker Noon’s one-week retrospective on the earlier GPT-6 Luna and Sol models captured real developer costs and shortcomings; the 6.1 refresh appears designed to close exactly those gaps in coding and document work.
The competitive implications reach beyond OpenAI. SpaceXAI’s Grok 4.6 debuted touting third-place performance on Artificial Analysis while costing less than half of what the prior Sol generation charged in standard mode — evidence that every major lab is now pricing against the same cost-per-intelligence curve. When benchmarks become commoditized, price becomes the battleground.
For buyers, the practical shift is that “flagship or bust” reasoning no longer holds. A model 1 Intelligence Index point behind the best available option, at less than a quarter of the task cost, changes procurement math for virtually every high-volume workload. Expect benchmark spreadsheets to gain a prominent cost column — if they had not already.
Should Teams Switch From Astra or Other Flagships?
For high-frequency, agentic workloads, the numbers argue for at least piloting GPT-6.1 Sol. A 1-point Intelligence Index deficit against a 75–80% cost reduction is a trade most economics teams will take, particularly since Artificial Analysis measured the advantage on Cost per Task — the metric that actually appears on invoices — rather than raw token price alone.
Fit depends on workload profile, and sources point to clear categories:
- Agentic coding — OpenAI and coverage from The New Stack both emphasize near-Astra performance on agentic coding benchmarks at one fifth the token price
- Computer use — the model’s upgrade brief from GPT-6 Sol explicitly targeted computer use capabilities
- Document understanding — improvements in dense document reading were highlighted in TechEBlog’s review of paying ChatGPT behavior
- Multi-step professional work — sources cite gains in code writing, debugging, and executing multi-step tasks
- Latency-sensitive apps — the 300 tokens-per-second Ultrafast tier targets interactive scenarios where flagship latency is a liability
Teams that should pause before switching are those pushing absolute frontier limits. If a workload sits in the narrow band of tasks where Astra outperforms Sol, that single Intelligence Index point may carry outsized value — frontier research, novel reasoning chains, and safety-critical judgments come to mind. The trade is only favorable when the capability delta is invisible to end users.
A pragmatic path: route production traffic to GPT-6.1 Sol, keep Astra as an escalation path for the small fraction of requests the mid-tier handles poorly, and measure the blended cost. Given that the model is already generally available on both the OpenAI API and Amazon Bedrock, running that experiment takes days, not quarters.
Frequently Asked Questions
What is GPT-6.1 Sol?
GPT-6.1 Sol is an upgraded version of GPT-6 Sol, announced by OpenAI at DevDay on September 29, 2026, with a focus on enhanced agentic coding, computer use, and professional work capabilities. It is available in the API at $2 per million input tokens and $10 per million output tokens, on Amazon Bedrock, and for paying ChatGPT users.
How much cheaper is GPT-6.1 Sol than GPT-6 Astra?
OpenAI says GPT-6.1 Sol delivers near-Astra intelligence at one fifth of the price, calling it the most cost-efficient model for its performance available today. Artificial Analysis’ independent framing puts it at 1 point below GPT-6 Astra in the Intelligence Index at less than one quarter of the Cost per Task.
Is GPT-6.1 Sol available on Amazon Bedrock?
Yes. An AWS Machine Learning blog post confirms GPT-6.1 Sol is generally available on Amazon Bedrock, bringing stronger reasoning to coding, computer use, and professional workloads that run frequently. This lets AWS-based enterprises access the model through existing Bedrock pipelines rather than integrating directly with OpenAI’s API.
Did GPT-6.1 Astra launch at the same time?
No. The Decoder reports that the planned flagship GPT-6.1 Astra is staying under wraps for now, while GPT-6.1 Sol launched on September 29, 2026. Astra appears in the announcement only as the benchmark OpenAI claims Sol nearly matches.
Summary
OpenAI’s GPT-6.1 Sol launch reshapes the middle of the model market. The key takeaways:
- Near-flagship at mid-tier pricing: 1 point below GPT-6 Astra on the Intelligence Index at one fifth the token price, per OpenAI, and less than a quarter of Astra’s Cost per Task per Artificial Analysis
- Speed as a differentiator: a new Ultrafast tier runs at 300 tokens per second, targeting interactive coding, support, and incident response
- Wide availability from day one: OpenAI API ($2/$10 per million input/output tokens), Amazon Bedrock general availability, and paying ChatGPT access
- Ruthless refresh cadence: GPT-6.1 Sol replaced GPT-6 Sol after just 7 days
- Astra remains unreleased: the flagship GPT-6.1 generation is still under wraps, making Sol the practical default for cost-sensitive teams
If your team runs high-volume agentic workloads, the blended-cost experiment is worth running this week. Check the AWS Bedrock post and OpenAI’s announcement, then route a slice of production traffic and compare invoices.
Further reading: TechCrunch’s launch coverage, VentureBeat on the Ultrafast tier, and unite.ai’s DevDay report.