Week 40 - complete

API costs and Copilot billing

September 28 to October 4, 2026 UTC. A new Sol model lowers cache-read costs, Astra adds an optional speed tier, and eligible Copilot teams move to upfront seat billing.

3 cost-choice and billing eventsOfficial sourcesReviewed October 5

What changed for your wallet

01
Sol cache reads cost less. GPT-6.1 Sol reads cached input at $0.10 per million tokens versus GPT-6 Sol at $0.20. A 50% lower cache-read rate, not a 50% reduction in your entire bill.
02
Astra speed has an optional premium. Ultrafast is 6x Standard token rates. Global short-context input/output are $60/$300 per million tokens versus $10/$50.
03
Copilot changes cash-flow timing. Existing Business and Enterprise customers paying by credit card or PayPal move to upfront charges at their next billing cycle starting October 1. Seat prices remain $19/$39 per month.

Cost and billing summary

Tool	What Changed	Was	Now	Impact	Who It Affects
GPT-6.1 Sol	New model launched September 29	GPT-6 Sol predecessor: $0.20 cache read	$0.10 cache read; $2 input / $10 output	Cache read -50%	API teams evaluating migration with repeated cached context
GPT-6 Astra	Optional Ultrafast tier added September 29	Standard comparison: $10 input / $50 output	Ultrafast: $60 input / $300 output	Optional 6x	API teams paying for lower latency
GitHub Copilot	Upfront timing takes effect from October 1 at next renewal	Unchanged seat prices: Business $19 / Enterprise $39	Same prices; assigned seats paid upfront at next cycle	Timing change	Existing Business/Enterprise customers paying by credit card or PayPal

API amounts are USD per million tokens for global processing with up to 272,000 input tokens. Copilot amounts are USD per granted seat per month. The Was column identifies a predecessor, a Standard comparison, or unchanged seat prices; it does not invent former GPT-6.1 Sol or Ultrafast prices.

Past briefings: Week 39 · Briefing archive

GPT-6.1 Sol: lower cache reads on a new model

OpenAI released GPT-6.1 Sol on September 29. Standard global pricing up to 272,000 input tokens is $2 input, $0.10 cache read, $2.50 cache write and $10 output per million tokens.

Its GPT-6 Sol predecessor has Standard rates of $2 input, $0.20 cache read, $2.50 cache write and $10 output in the same context bracket. Only the cache-read unit rate is 50% lower. This is a new model, not a retroactive GPT-6 Sol repricing. Total cost depends on cache reuse, token counts and the model's results on your tasks.

Above 272,000 input tokens, the full request uses higher rates: $4 input, $0.20 cache read, $5 cache write and $15 output per million tokens. Regional processing and alternative processing tiers have separate premiums or discounts. These API rates do not establish any ChatGPT subscription price or allowance change.

Who This Affects

API developers considering migration from GPT-6 Sol, especially with repeated cached context. Compare task quality and total token usage before switching. See the OpenAI API rate table.

Official sources: OpenAI API changelog (September 29 launch and September 22 predecessor entry), API pricing and GPT-6.1 Sol model documentation. Retained October 5, 2026 captures support the comparison, cache-write rate and full-request context rule.

Astra Ultrafast: choose speed at a higher token rate

On September 29, OpenAI added optional GPT-6 Astra Ultrafast to the Responses API. Standard global short-context rates are $10 input, $1 cache read, $12.50 cache write and $50 output per million tokens. Ultrafast rates are $60 input, $6 cache read, $75 cache write and $300 output.

All four token rates are 6x Standard. This is an optional tier, not a compulsory hike to Standard pricing. Enable it per request with model: "gpt-6-astra" and service_tier: "ultrafast". Above 272,000 input tokens, the full request uses $120 input, $12 cache read, $150 cache write and $450 output per million tokens.

Ultrafast supports global processing and US data residency only. EU and other regional inference residency are unsupported. Availability is subject to rate limits. The amounts here cover global processing; supported regional endpoints have a separate uplift. These API rates are separate from ChatGPT allowances.

Who This Affects

Developers with latency-sensitive workloads. Test whether the time saved justifies the token premium, check residency requirements and rate limits, and keep Standard for tasks that do not need Ultrafast. Compare the API tiers and rates.

Official sources: OpenAI API changelog (September 29), API pricing and Ultrafast guide. Retained October 5, 2026 captures establish prices, request settings, rate limits and residency restrictions.

Copilot: budget for upfront seats at the next cycle

Existing Business and Enterprise customers paying by credit card or PayPal move to upfront assigned-seat charges starting October 1, 2026, at the start of their next billing cycle. This means not everyone was charged on October 1. Business remains $19 and Enterprise $39 per granted seat per month in USD.

GitHub announced this on August 28 and reiterated it on September 3. It is an effective-date story, not announced this week. New seat assignments require payment before access. Mid-cycle additions remain prorated from assignment to the end of the billing cycle; seat removal produces no prorated refund and is reflected in the next monthly cycle.

The announcement's payment-method scope does not establish this change for invoiced enterprises or individual plans. Exact prior billing timing is not established by the retained evidence, so no earlier payment schedule is claimed here.

Who This Affects

Organization billing owners on Business or Enterprise who pay by credit card or PayPal. Review assigned seats and the next renewal date, then budget for upfront payment. The Copilot pricing guide preserves the unchanged seat prices.

Official sources: GitHub's August 28 billing announcement, September 3 follow-up and Copilot plan documentation. Retained October 5, 2026 captures establish eligibility, next-cycle timing, unchanged seat prices and proration rules.

Switch Watch

Three practical actions based on this week's verified cost and billing events.

Cached context

Test Sol before migrating

Compare GPT-6.1 Sol with GPT-6 Sol on real tasks. Lower cache-read rates help only to the extent that your workload reuses cached input.

Latency budget

Use Ultrafast selectively

Measure response time and total token cost. Keep Standard for tasks whose latency does not justify the optional 6x rate.

Seat renewal

Review Copilot assignments

Check your payment method, next billing date and assigned seats. This timing change gives no evidence of a seat-price increase.

OpenAI API costs Copilot seat costs Compare subscription plans

Three verified cost-choice and billing events, not three sticker-price changes. Coverage is the complete UTC reporting week. API billing and subscription allowances are separate products; no subscription allowance changes are inferred here.