The Revenue AI Report
← Reality Check

GPT-6.1 Sol: Near-Astra Work at One-Fifth Price

gpt-6.1-sol Standard short list is $2/$10, one-fifth of Astra $10/$50. Cache read $0.10, cache write $2.50. Fast is 2x, Batch/Flex is 50%, input over 272K rebills the full request, regional is +10% where eligible. Sol Ultrafast is now live as of Oct 8 2026. See [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form). Tag vendor per-task dollars VENDOR / UNPROVEN. Name the default model ID and the accepted-outcome owner before volume agents point at Sol.

Jonathan Kvarfordt · Published September 29, 2026 · Last reviewed October 13, 2026 · 8 min read

Why trust this analysis?

The short answer

If Sol list is about one-fifth of Astra, which model ID is our Order Form default for volume agents, which meters move the bill off the sticker, and who owns accepted outcome before we point volume at gpt-6.1-sol?

API model id is gpt-6.1-sol. Standard short list (input at or under 272K) is $2/MTok in and $10/MTok out, one-fifth of GPT-6 Astra $10/$50. Cached input $0.10 (5% of uncached input; half of prior GPT-6 Sol cached input $0.20). Cache write $2.50 (1.25x uncached input). Astra short cache is $1.00 read / $12.50 write. Input over 272K bills the full request at long Standard: input $4, cached input $0.20, cache write $5, output $15 (2x input and cache, 1.5x output). Fast is 2x Standard: short Fast $4 / $0.20 / $5 / $20; long Fast $8 / $0.40 / $10 / $30. service_tier is fast or priority (priority renamed Fast on 30 Jul 2026). Fast is unavailable with EU data residency. Batch and Flex are 50% of the Standard rate that applies. Regional processing is +10% where eligible. Sol Ultrafast is now live as of Oct 8 2026. See [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form). Near-Astra and per-task cost claims are VENDOR / UNPROVEN, including Terminal-Bench Science 0.1 at max effort (Sol about $5.47, Astra about $23.80, Opus 5.5 about $23.21 per task; vendor scores Sol 57.0% vs Astra 68.1%). Name gpt-6.1-sol on the Order Form. Leave the Bedrock inference profile blank until Bedrock docs name it. Name the accepted-outcome owner (role may stay blank on the draft; do not invent a person) before volume agents point at Sol. List fifth is not bill fifth and is not the task ladder.

Decision rule

Refuse an unnamed default. Lock the Sol vs Astra list fifth on the paper, tag every per-task bench VENDOR / UNPROVEN, map effort, cache, Fast, Batch/Flex, over-272K, and regional meters, review Sol Ultrafast, now live as of Oct 8 2026, on [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form), name the accepted-outcome owner or flag Proof Gap, and do not invent a Bedrock profile or a forever-fifth SKU.

Evidence

  • developers.openai.com/api/docs/models/gpt-6.1-sol (29 Sep 2026): model id gpt-6.1-sol; Standard short $2/$10; cached input $0.10; cache write $2.50; context 1,050,000; max output 128,000; effort low/medium (default)/high/xhigh/max; none and minimal not supported
  • developers.openai.com/api/docs/models/gpt-6-astra: Astra Standard short $10/$50; cached input $1.00; cache write $12.50. Sol list is one-fifth of Astra list on input, output, and cache write; cached input is one-tenth
  • developers.openai.com/api/docs/pricing and openai.com/api-fast-mode: over 272K full-request long Standard $4 / $0.20 / $5 / $15; Fast 2x (short $4/$0.20/$5/$20, long $8/$0.40/$10/$30); Batch/Flex 50%; regional +10% where eligible; Fast unavailable with EU data residency; Sol Ultrafast is now live as of Oct 8 2026. See [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form)
  • AWS Bedrock post (29 Sep 2026) and TNW: Bedrock GA same day; DeepSWE / OSWorld / Terminal-Bench / AutomationBench per-task claims. Tag VENDOR / UNPROVEN. Do not invent the Bedrock inference profile string
  • Vendor Terminal-Bench Science 0.1, max effort, OpenAI preliminary: Sol about $5.47/task vs Astra about $23.80 vs Opus 5.5 about $23.21; vendor scores Sol 57.0% vs Astra 68.1%. Not an Order Form receipt

Operator action

Today: put gpt-6.1-sol on the Order Form as the API default, lock list and multiplier lines, review Sol Ultrafast, now live as of Oct 8 2026, on [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form), tag vendor per-task dollars VENDOR / UNPROVEN, name the chargeback owner for effort, cache, Fast, Batch/Flex, long context, and regional, leave accepted outcome blank until a role is named, and do not invent a Bedrock profile or a forever-fifth SKU.

Last reviewed

GPT-6.1 Sol is near-Astra agent work at about one-fifth Astra list $/MTok. Which model ID is the Order Form default?

List sticker math vs Astra, the meters that move the bill after the sticker, and who owns the API default before volume agents point at Sol.

The argument

How this reality check breaks down

A map of the sections ahead, in the order the case is made. The sourced charts are in the research library.

Contents diagram for GPT-6.1 Sol: Near-Astra Work at One-Fifth Price, listing the sections: GPT-6.1 Sol Order Form one-fifth Astra list, Gpt-6.1-sol $2 $10 vs Astra $10 $50, Sol cache $0.10 write $2.50 Fast 2x long context, Terminal-Bench per-task VENDOR UNPROVEN.

Model ID for the API is `gpt-6.1-sol`. Standard short list is $2 / $10 per 1M input / output. That is the same Sol-tier sticker as GPT-6 Sol, and one-fifth of GPT-6 Astra $10 / $50. Near-Astra quality and per-task "fifth" claims are VENDOR / UNPROVEN until your eval. Name the default ID on the Order Form and the chargeback owner for effort, cache, Fast, Batch/Flex, and the over-272K band before you make Sol the volume default (GPT-6.1 Sol; GPT-6 Astra).

List sticker (locked)

Standard short context (input at or under 272K), $/MTok:

Reality Check

GPT-6.1 Sol: Near-Astra Work at One-Fifth Price

Source-cited charts live in the research library.

GPT-6.1 Sol: Near-Astra Work at One-Fifth Price. Diagram showing Evidence, Analysis, Decision, Outcome.
  • Input: Sol $2.00. Astra $10.00. Ratio 1/5.
  • Output: Sol $10.00. Astra $50.00. Ratio 1/5.
  • Cached input: Sol $0.10. Astra $1.00. Ratio 1/10. Sol cached input is 5% of uncached input, and half of prior GPT-6 Sol cached input $0.20.
  • Cache write: Sol $2.50. Astra $12.50. Ratio 1/5. Sol cache write is 1.25x uncached input.

The fifth-price frame is token-sticker math. It is not the task ladder. This cite is not a token-cut decoder. That lives on GPT-6 / Opus 5.5 token-cut Order Form.

Side-by-side Standard short $/MTok stickers — Sol $2/$10 with cache $0.10 read / $2.50 write vs Astra $10/$50 with cache $1.00 / $12.50; fifth-price frame callout that list is not the task ladder
Standard short list. Sol $2/$10 is one-fifth of Astra $10/$50. List is not the task ladder.

Meters that leave the sticker

Also locked on Sol docs and the pricing page (pricing; Fast mode):

  • Context: 1,050,000. Max output 128,000. Knowledge cutoff Apr 30 (soft).
  • Effort: `low` / `medium` (default) / `high` / `xhigh` / `max`. `none` and `minimal` are not supported. Default medium. `max` and `xhigh` stretch tokens per task.
  • Long context (input over 272K): 2x input and cache rates, 1.5x output, on the full request. Long Standard: input $4.00, cached input $0.20, cache write $5.00, output $15.00.
  • Fast: 2x Standard. `service_tier` `fast` or `priority` (priority renamed Fast on 30 Jul 2026). Short Fast: $4 / $0.20 / $5 / $20. Long Fast: $8 / $0.40 / $10 / $30. Fast is unavailable with EU data residency.
  • Batch / Flex: 50% of the Standard rate that applies (short or long). Not the hero line.
  • Regional processing: +10% where eligible.
  • Ultrafast: now live as of Oct 8 2026. See Sol Ultrafast Order Form decoder. Seat packaging for Ultrafast stays on OpenAI Pro 200 / Pro 500 Ultrafast.
Four-panel multiplier strip — Fast 2× with short/long dollars; Batch/Flex 50%; long-context over 272K full-request rates; regional +10% and Sol Ultrafast now live as of Oct 8 2026; see [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form)
Fast 2x, Batch/Flex 50%, over-272K full request, regional +10%. Sol Ultrafast is now live as of Oct 8 2026. See [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form).

Vendor per-task claims (VENDOR / UNPROVEN)

OpenAI, AWS, and secondary coverage relay OpenAI's own benches. Tag every line VENDOR / UNPROVEN. Do not put them on the Order Form as receipt math. The figure below is Terminal-Bench Science 0.1 only, and it is vendor-reported.

  • Terminal-Bench Science 0.1, max effort (OpenAI preliminary): Sol about $5.47 / task vs Astra about $23.80 vs Opus 5.5 about $23.21. Vendor scores: Sol 57.0% vs Astra 68.1%. OpenAI still points Astra at the hardest science. Not an Order Form receipt.
  • DeepSWE v1.1 (vendor, AWS / TNW): Sol matches Astra at roughly one-fifth cost per task, and beats GPT-6 Sol best by 6.4 points.
  • OSWorld 2.0 (vendor, TNW): within about 2.1 points of Astra at about one-seventh cost per task.
  • AutomationBench (vendor, TNW; medium-vs-medium in secondary coverage): Sol +2.2 vs Opus 5.5 at about one-third cost.
One-row VENDOR-REPORTED per-task cards — Sol $5.47 vs Astra $23.80 vs Opus 5.5 $23.21 on Terminal-Bench Science 0.1 max effort; takeaway that sticker fifth is not the bake-off receipt
VENDOR / UNPROVEN. Terminal-Bench Science 0.1 max effort. Sticker fifth is not the bake-off receipt.

Bedrock and where Sol shows up

Amazon Bedrock: GPT-6.1 Sol GA 29 Sep 2026. AWS says near-Astra for everyday agent work and relays the DeepSWE match at about one-fifth cost per task (VENDOR / UNPROVEN). Inference on Bedrock. AWS says data is not used for OpenAI training, and makes zero-operator access claims in that post. Exact Bedrock inference profile string: do not invent it. Pull it from Bedrock docs before the Order Form names a cloud SKU. API model id is `gpt-6.1-sol` (AWS Bedrock post).

TNW (secondary): ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu. Not yet in the main ChatGPT chat. Do not treat that as an Order Form line (TNW; deployment note).

Chargeback lines that move the bill off the sticker

Pull these meters. The press "fifth" dies when they fire:

  1. Model ID / serving path. `gpt-6.1-sol` vs `gpt-6-astra`, or a Bedrock profile once docs name it. The Order Form default must be named. Do not invent the profile string.
  2. Effort. Default `medium`. `max` / `xhigh` stretch tokens per task.
  3. Cache write vs cached input. Write $2.50. Hit $0.10 on short Standard. Miss ratio on agent loops matters. Long-context cache is $5.00 write / $0.20 read, on the full request.
  4. Fast (2x). Short and long Fast dollars above. Not Ultrafast. Sol Ultrafast is now live as of Oct 8 2026. See Sol Ultrafast Order Form decoder.
  5. Batch / Flex (50%) vs real-time Standard.
  6. Over 272K. 2x input and cache, 1.5x output, on the full request.
  7. Regional +10% where used. Fast does not run with EU data residency.
  8. Accepted-outcome denominator. Cost per successful agent task, not list $/MTok alone. Name the role that decides Sol accepted vs escalate to Astra (or Opus). A blank name on the draft is allowed. Do not invent a person. Blank at go-live is a Proof Gap.
Chargeback decision tree — model id × tier × short vs over-272K × cache write vs read × regional uplift, plus blank accepted-outcome owner name/role chips before volume
Default model id, then tier, context band, cache, and regional. Accepted-outcome owner stays blank until you name the role.

Soft opens (do not invent)

  • Exact Bedrock model ID / inference profile until Bedrock docs cite it.
  • Sol Ultrafast is now live as of Oct 8 2026. See Sol Ultrafast Order Form decoder. Fast remains 2x Standard.
  • Fortune-500 adoption headcounts. Any Salesforce quote in secondary coverage is vendor-only. Not your proof.
  • No Order Form SKU, pack, or commitment that says "1/5 forever."

Distinct from live cites

Order Form / chargeback checklist

  1. Default model ID: `gpt-6.1-sol` on the API. Bedrock profile only after docs name it.
  2. List: $2 / $10. Cache read $0.10. Cache write $2.50. Astra compare $10 / $50 is 1/5 list math (locked). Per-task fifth or seventh is VENDOR / UNPROVEN.
  3. Multipliers: Fast 2x (short and long dollars). Batch/Flex 50%. Over 272K full-request rates. Regional +10%. Sol Ultrafast is now live as of Oct 8 2026. See Sol Ultrafast Order Form decoder.
  4. Chargeback owner named for effort, cache, Fast, long context, Batch/Flex, and regional.
  5. Accepted-outcome blank. Name the role. If missing at go-live, Proof Gap.
  6. Soft subscribe only.

Proof Gap

  • List 1/5 is not bill 1/5 once Fast, cache write, over-272K, or max effort fire.
  • Vendor DeepSWE, OSWorld, Terminal-Bench, and AutomationBench are not your accepted outcome.
  • An unnamed API or Bedrock default is unforeseeable chargeback.
  • A missing accepted-outcome owner means you cannot tell Sol-accepted from escalate-to-Astra.
  • No forever-fifth commitment sits on this paper.

Who reviews which line (Eight-Seat lite)

  • RevOps / Finance: list fifth vs task-cost claim; cache, Fast, Batch/Flex, long context, regional; accepted-outcome blank.
  • Platform: API model id `gpt-6.1-sol`; effort default; cache policy; Fast vs Batch routing; EU residency vs Fast. Bedrock profile only when docs name it.
  • Enablement: do not train the press fifth as the bill. Gate volume agents until the default ID and the accepted-outcome role are named.

Sources

  1. OpenAI, GPT-6.1 Sol model docs: developers.openai.com/api/docs/models/gpt-6.1-sol
  2. OpenAI, GPT-6 Astra model docs: developers.openai.com/api/docs/models/gpt-6-astra
  3. OpenAI API pricing: developers.openai.com/api/docs/pricing
  4. OpenAI, API Fast mode: openai.com/api-fast-mode
  5. AWS, GPT-6.1 Sol on Amazon Bedrock (29 Sep 2026): aws.amazon.com Bedrock post
  6. TNW, secondary: thenextweb.com GPT-6.1 Sol price
  7. OpenAI deployment safety note: deploymentsafety.openai.com/gpt-6-1-sol/introduction

Soft CTA

New research lands in the weekly issue first. Subscribe

Take it to the room

The short list from this piece

Pulled from the argument above, written so you can read it out in a pipeline or board review.

Checklist diagram summarising GPT-6.1 Sol: Near-Astra Work at One-Fifth Price: OpenAI, GPT-6.1 Sol model docs; OpenAI, GPT-6 Astra model docs; OpenAI API pricing; OpenAI, API Fast mode: openai.com/api-fast-mode; AWS, GPT-6.1 Sol on Amazon Bedrock (29 Sep 2026).

Frequently asked questions

Is Sol one-fifth the price of Astra?
On the Standard short list sticker, yes. $2 / $10 vs $10 / $50. Cache write is also 1/5 ($2.50 vs $12.50). Cached input is 1/10 ($0.10 vs $1.00). That is list math, not cost per accepted task.
Which meters move the bill off the fifth?
Effort (default medium), cache write vs cached read, Fast 2x, Batch/Flex 50% vs real-time, the over-272K full-request band, and regional +10%. The denominator is accepted outcome, not list $/MTok alone.
Is Ultrafast priced for gpt-6.1-sol?
Yes. As of Oct 8 2026, short Sol Ultrafast list is $12 / $0.60 / $15 / $60 per million tokens for input / cached input / cache write / output. Fast remains 2x Standard. See [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form). Seat packaging stays on [Pro 200 / Pro 500 Ultrafast](/blog/openai-pro-200-usage-cut-pro-500-ultrafast-order-form).
What Bedrock string do we print on the Order Form?
Do not invent one. Bedrock GA is 29 Sep 2026. The API model id to name today is gpt-6.1-sol. Pull the inference profile from Bedrock docs before it hits the paper.
Can vendor per-task dollars go on the Order Form?
No. DeepSWE, OSWorld, Terminal-Bench Science 0.1, and AutomationBench are VENDOR / UNPROVEN until your eval. Terminal-Bench max-effort dollars (about $5.47 vs $23.80 vs $23.21) are not a receipt. Sticker fifth is not the bake-off.
Who owns accepted outcome?
The role that decides Sol accepted vs escalate to Astra (or Opus), before volume. A blank name is fine on the draft. Do not invent a person. Still blank at go-live is a Proof Gap.
What must be on the Order Form today?
Default id gpt-6.1-sol. List, cache, Fast, Batch/Flex, long-context, and regional lines. Sol Ultrafast is now live as of Oct 8 2026. See [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form). VENDOR / UNPROVEN on every per-task bench. Chargeback owner. Accepted-outcome blank. No forever-fifth SKU. No invented Bedrock profile.

Share this page

Posting to Instagram or TikTok? Copy the link, it carries the title, summary and share image.

Subscribe

Get the next Reality Check before you sign the order form.

Arrives weekly by email. Free. Unsubscribe anytime. By subscribing you agree to our Privacy policy and Terms. We never sell or share the list.

Keep reading