OpenAI’s DevDay 2026 keynote had a lot of shiny new things, from always-on Dots agents to a new Pro 500 plan, but the most consequential announcement for anyone who pays for tokens was also the quietest one. GPT-6.1 Sol is a direct upgrade to GPT-6 Sol, the model OpenAI launched just a week earlier, and the company says it nearly matches GPT-6 Astra on agentic coding, computer use, and professional work at roughly one-fifth of Astra’s token prices.
The price is what makes this interesting. GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens, exactly the same API price as GPT-6 Sol. The “.1” is a capability upgrade, not a price hike, and cached input just got dramatically cheaper.
GPT-6.1 Sol vs GPT-6 Sol: what the name change actually means
If you are confused by the naming, the history helps. OpenAI launched GPT-6 Sol on September 22 at $2 per million input tokens and $10 per million output tokens, alongside the cheaper GPT-6 Luna ($0.10/$0.50). GPT-6.1 Sol, released September 29, keeps the same $2/$10 pricing while improving scores across nearly every benchmark OpenAI reported. Think of it as a free mid-cycle refresh rather than a new tier.
The standout pricing change is cached input: $0.10 per million tokens. OpenAI describes that as 95% below standard input pricing and 50% below GPT-6 Sol’s cached input price. At today’s rate of roughly ₹96 to the dollar, the API price works out to about ₹192 per million input tokens, ₹960 per million output tokens, and cached input at roughly ₹9.60 per million. For Indian developers building agent workloads with repeated context, the cache pricing alone is a meaningful cut.
Price and benchmark snapshot: GPT-6.1 Sol vs the field
| Model | Input / M tokens | Output / M tokens | Cached input / M | Terminal-Bench Science cost per task |
|---|---|---|---|---|
| GPT-6.1 Sol (new, Sept 29) | $2 | $10 | $0.10 | $5.47 |
| GPT-6 Sol (Sept 22) | $2 | $10 | ~2x the new rate (OpenAI says the new cached price is 50% lower) | not disclosed |
| GPT-6 Astra (flagship) | roughly 5x Sol’s prices, per OpenAI’s “one-fifth” claim | not disclosed | $23.80 (top score 68.1%) | |
| Claude Opus 5.5 | not disclosed; Anthropic cut Opus 5.5 pricing ~40% vs its predecessor | $23.21 | ||
Benchmarks: where GPT-6.1 Sol chases GPT-6 Astra
OpenAI’s reported results, from its own announcement and the Unite.AI rundown, paint a consistent picture: near-flagship quality at working-model prices.
On DeepSWE v1.1, a benchmark for complex software engineering in real codebases, GPT-6.1 Sol matches Astra at roughly one-fifth of the cost and beats GPT-6 Sol’s best score by 6.4 percentage points, at lower reasoning effort. On OSWorld 2.0’s offline computer-use workflows, it outperforms GPT-6 Sol by 7 percentage points at maximum reasoning effort and lands within 2.1 percentage points of Astra’s score at roughly one-seventh of the cost per task. On GDP.pdf, which tests answering professional questions from complex PDF documents, it scores higher than Opus 5.5 with fallbacks at less than half the cost per task, and approaches Astra at roughly one-fifth of the cost.
The most striking number is on Terminal-Bench Science 0.1, covering scientific workflows from data analysis to theorem proving. GPT-6.1 Sol more than doubles GPT-6 Sol’s score while averaging $5.47 per task, against $23.21 for Opus 5.5 and $23.80 for Astra. Astra still holds the highest score of the group at 68.1%, so the flagship keeps its crown, but the price per task gap is roughly 4x.
On AutomationBench, which tests multi-step business workflows, the model scores 2.2 percentage points above Opus 5.5 at medium reasoning effort for roughly a third of the cost, and 4.8 points above GPT-6 Sol at the same setting. Factuality improved too: the share of responses with a factual error fell from 11.4% to 7.7%, about a 32% reduction, and the error rate stays within 1.9 percentage points of Astra’s across tested settings. OpenAI also reports better alignment results, including failures to disclose a broken search tool dropping from 4.9% (GPT-6 Sol) to 2.1%, close to Astra’s 1.5%, with no observed attempts to bypass an automated safety reviewer.
For comparison with the rival camp, Claude Opus 5.5 arrived with its own ~40% price cut and strong coding scores, so this is becoming a price war fought benchmark by benchmark.
Where GPT-6.1 Sol is available, and where it is not
From September 29, the model is available to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, and developers can call it in the API as gpt-6.1-sol. One notable gap: it is not yet available in ChatGPT Chat, so consumer chat users will wait.
The same keynote introduced Ultrafast, a premium speed tier promising up to 8x faster token generation (300 tokens per second) in Codex and up to 6x in the API. Astra Ultrafast is available now in the API and on Pro 500 and Enterprise plans, while a GPT-6.1 Sol Ultrafast option follows in the coming days. OpenAI also launched a new Pro 500 plan with what it describes as 25x the ChatGPT Plus usage allowance plus Ultrafast access, aimed at power users and teams running agents heavily.
The Astra-shaped hole in the lineup
Notably absent from DevDay was GPT-6.1 Astra. OpenAI halted that release one day before the keynote over safety concerns, with its head of safety systems telling Barron’s the model “didn’t quite meet the bar” on staying within scope and authorization. Astra remains the flagship, and Barron’s reports the new always-on Dots agents are powered by it, but the scrapped 6.1 upgrade leaves GPT-6.1 Sol as the newest frontier model OpenAI will actually sell you. In that light, Sol 6.1 closing the gap on Astra matters more than it might look: it may be the best OpenAI model available at scale for a while.
What this means for developers and buyers
If you already build on the OpenAI API, the move is simple: switch gpt-6-sol to gpt-6.1-sol. Same price, better benchmark scores, and a much cheaper cache. Cache-heavy agent workloads, the kind that replay the same documents and tool definitions across many tasks, benefit the most, since cached input at $0.10 per million is now 20x cheaper than uncached input.
If you were deciding between OpenAI and Anthropic for coding agents, GPT-6.1 Sol’s $5.47-per-task figure on Terminal-Bench Science vs $23.21 for Opus 5.5 is the number to watch, though remember Astra still outscores both and each vendor benchmarks its own models. And if you have been waiting for Astra-class quality at a working-model price, this is the closest OpenAI’s official lineup gets for now, especially with the 6.1 Astra upgrade on ice.
Our DevDay preview expected agent safety and new model tiers to dominate the keynote, and the safety story did shape the event, just in the form of a release OpenAI chose not to ship.
