OpenAI's $500 plan and GPT-6.1 Sol: speed is the new premium
OpenAI now sells a $500-a-month ChatGPT plan built around raw speed, while its new mid-range model gets close to the flagship for a fifth of the price. Here is what changed, what it costs, and who should care.

Key points
- At DevDay on 29 September, OpenAI launched GPT-6.1 Sol. It costs the same as GPT-6 Sol, $2 per million input tokens and $10 per million output tokens, but repeated (cached) input now costs half as much.
- OpenAI says 6.1 Sol comes close to its flagship GPT-6 Astra on coding and computer-use tests, at a fifth of Astra's price.
- A new ChatGPT Pro 500 plan costs $500 a month. It offers about 25 times the usage of the $20 Plus plan and an 'Ultrafast' mode for Astra of up to 300 tokens a second.
- The $200 Pro plan loses half its Codex and ChatGPT Work allowance at the end of October, nudging heavy users towards the new tier.
- For most businesses the cheaper model matters more than the faster plan: speed costs six times as much per token.
OpenAI’s developer conference on 29 September produced two announcements that pull in opposite directions. One makes good AI cheaper: a new mid-range model that gets close to the flagship. The other makes the best AI faster, for a lot more money: a $500-a-month ChatGPT plan. Together they show where the AI business is heading.
GPT-6.1 Sol: nearly flagship, a fifth of the price
GPT-6.1 Sol replaces GPT-6 Sol, which launched only recently to mixed reviews. The list price is unchanged, $2 per million input tokens and $10 per million output tokens, but cached input, text the model has already seen in an earlier request, now costs $0.10 per million instead of $0.20. That matters for chatbots and coding agents, which resend the same long instructions and documents again and again.
OpenAI says the new model comes close to GPT-6 Astra, its most capable model, on several tests of coding and operating a computer. Independent write-ups of the published results show it well ahead of GPT-6 Sol on security and troubleshooting benchmarks, though still behind Astra. OpenAI also reports fewer factual errors: 7.7% at low reasoning effort, down from 11.4%.
Astra costs five times as much: $10 per million input tokens and $50 for output.
What a real workload costs
Prices per million tokens are hard to picture, so here is a typical coding-agent job: 1,000 steps, each sending 40,000 tokens of context, 80% of it repeated from earlier steps, and writing 2,500 tokens back.
The same agent job on four OpenAI options
Cost in US dollars of 1,000 agent steps, each with 40,000 input tokens (80% cached) and 2,500 output tokens, at OpenAI's standard list prices
Source: AI Beat calculation from OpenAI API pricing, 30 September 2026
The cheaper cache saves only a few percent on this job because output dominates the bill; for workloads that reread large documents and give short answers, it can reach about a third. The bigger jump is between models: Astra costs about five times as much as Sol, and its Ultrafast mode six times as much again.
You can run your own numbers in our AI API cost calculator, which now includes GPT-6.1 Sol.
Pro 500: paying for speed
The new ChatGPT plan, called Pro 500, costs $500 a month. OpenAI’s line-up of paid personal plans now runs from Go and Plus to three Pro tiers at $100, $200 and $500.
ChatGPT's personal plans, monthly price
US dollars per month. Business and Enterprise plans are priced per seat.
Source: OpenAI help centre and pricing as reported by Engadget and others, September 2026
What the extra money buys:
- About 25 times the usage of Plus, according to OpenAI’s launch materials as reported by several outlets. OpenAI does not publish exact limits.
- Ultrafast mode for GPT-6 Astra in ChatGPT Work and the Codex coding tool, at up to 300 tokens a second, several times faster than normal. Ultrafast uses up the allowance eight times faster, so heavy use of it quickly shrinks that 25x.
- Dot, a personal agent, whose conversations don’t count towards limits.
At the same time, the $200 Pro plan’s allowance for Codex and ChatGPT Work drops from 20 times Plus to 10 times at the end of October, and its weekly GPT-6 Pro messages from 200 to 100. Existing subscribers get a one-off credit.
On the hardware: posts on social media claimed the Astra Ultrafast mode runs on chips from Cerebras. That is not confirmed. Cerebras has announced that it powers a fast mode for the older GPT-5.6 Sol, and the analysis firm SemiAnalysis reports that 6.1 Sol’s fast mode runs on Nvidia GPUs. Neither OpenAI nor Cerebras has said what runs Astra Ultrafast.
Who should care
- Most businesses using the API: switch from GPT-6 Sol to 6.1 Sol. Same price, better results, cheaper caching.
- Teams choosing between Sol and Astra: test 6.1 Sol first. If it is close enough on your tasks, the saving is about 80%.
- Individual power users: Pro 500 makes sense only if waiting for the model is the bottleneck, for example developers running long coding sessions all day. For everyone else, Plus or Pro 100 is still the better buy.
- Everyone: watch the direction of travel. Capability is getting cheaper, while speed and volume become the premium products. Expect other AI companies to price the same way.
Sources
- OpenAI API: Pricing
- OpenAI API: GPT-6.1 Sol model page
- OpenAI: Introducing GPT-6.1 Sol (29 September 2026)
- TechCrunch: OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astra and costs less (29 September 2026)
- VentureBeat: GPT-6.1 Sol offers Astra-like performance at a fifth of the price (29 September 2026)
- DataCamp: GPT-6.1 Sol benchmarks
- Engadget: OpenAI adds $500 Pro subscription, reduces its $200 tier
- The Decoder (Mixed News): ChatGPT Pro 500 and lower Pro 200 allowance
- Cerebras: Cerebras powers Ultrafast mode for OpenAI's GPT-5.6 Sol (13 August 2026)
- OfficeChai: GPT-6.1 Sol Ultrafast runs on Nvidia GPUs, says SemiAnalysis (30 September 2026)
Comments
What do you think? No account needed. Please be kind; comments that attack people are removed.
No comments yet. Be the first.