Record reviewed readings of the week-43 and week-44 briefs.
Week 43 keeps the Kimi K3.1 Pro catalog row and the Claude Duet navigation label. Week 44 keeps GPT-6.1 Terra, the 29 October Mistral Large 4 fields, and the 30 October Max and Ultra plan text. Assistant: grok Assistant-Session: 01a1925d-3496-7011-a33e-6bfb99f4f4d3
This commit is contained in:
parent
c4f71a1231
commit
851c6a74ca
9 changed files with 110 additions and 22 deletions
18
SCOPE.md
18
SCOPE.md
|
|
@ -12,7 +12,8 @@
|
|||
- Frozen recommendations, deterministic historical replay and Markdown reports.
|
||||
- Reviewed publication records. An included claim must name a capture already in
|
||||
the ledger. A generated draft is not that record. Reviewed publications now
|
||||
cover the week-40, week-41, and week-42 briefs.
|
||||
cover the week-40 through week-43 briefs and the week-44 daily briefs through
|
||||
30 October.
|
||||
- Configured Europe/Berlin daily/weekly job runner, retries, locking, health receipts,
|
||||
and user systemd timer units for the local workstation.
|
||||
- Retention of unarchived ChatGPT and Grok automation briefs from the local drop folder,
|
||||
|
|
@ -37,9 +38,18 @@ prices and OpenRouter rows for Sonnet 5.5 and GPT-6.1 Sol. Week 41 publication
|
|||
and the OpenRouter rows for Haiku 5.5 and Mistral Large 4. Week 42 publication
|
||||
`db9e3a7068329a605daee94cc0429706d0aedfaeb42c4f167b88d21483783339` keeps the
|
||||
OpenRouter rows for Muse Wonder, Gemini 4 Argon, Muse Spark Open, and DeepSeek
|
||||
V4.1 Flash 1017, and the 18 October Haiku 5.5 and Sonnet 5.5 rates. The rest of
|
||||
each Grok brief stays excluded. Price readings are in `frontier/readings-2026-W40.md`,
|
||||
`frontier/readings-2026-W41.md`, and `frontier/readings-2026-W42.md`. A reviewed two-profile comparison is retained, and
|
||||
V4.1 Flash 1017, and the 18 October Haiku 5.5 and Sonnet 5.5 rates. Week 43 publication
|
||||
`2181566caba12b0325c604b53168a6c02acaeddcc88cefaf3963d1a46ee60f23` keeps the
|
||||
OpenRouter row for Kimi K3.1 Pro and the 25 October Opus 5.5, Sonnet 5.5, and
|
||||
Haiku 5.5 rates. The pricing page lists Max 5x and Max 20x through 25 October
|
||||
and adds the navigation label Claude Duet on 21 October. Week 44 publication
|
||||
`43113feeff26834ad3d7585469141fe2c13a4652b7c650500d4e4e23afba1630` keeps the
|
||||
OpenRouter rows for GPT-6.1 Terra and the 29 October Mistral Large 4 fields,
|
||||
and the 30 October Max and Ultra plan text. No week-44 weekly brief is retained,
|
||||
and there is no 31 October capture. The rest of each Grok brief stays excluded.
|
||||
Price readings are in `frontier/readings-2026-W40.md`, `frontier/readings-2026-W41.md`,
|
||||
`frontier/readings-2026-W42.md`, `frontier/readings-2026-W43.md`, and
|
||||
`frontier/readings-2026-W44.md`. A reviewed two-profile comparison is retained, and
|
||||
it withholds Frontier, Efficient, and Budget winners because requester access and
|
||||
exact benchmark results are not evidenced.
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1 @@
|
|||
{"claims":[{"capture_id":"0dfc47ac722749067c30845c598b72023ebbabc0793b386bdac961bd4e65fa28","disposition":"included","statement":"The 22 October OpenRouter models capture does not list moonshotai/kimi-k3.1-pro."},{"capture_id":"252f74fe0aa7a965a8a875817768a9c1bdda4b9b765889c985080e230230c8cd","disposition":"included","statement":"The 23 October OpenRouter models capture lists moonshotai/kimi-k3.1-pro as MoonshotAI: Kimi K3.1 Pro, context 1048576, prompt 0.000002 and completion 0.00001 per token. The row does not state a percentage price cut, a parameter count, a weights-release date, or an index score."},{"capture_id":"97562814bb3287f7d90aac8e61bc8ae1d3e72c29d5c41438c65a22ca8c9c56c0","disposition":"included","statement":"The 25 October OpenRouter models capture lists moonshotai/kimi-k3.1-pro as MoonshotAI: Kimi K3.1 Pro, context 1048576, prompt 0.000002 and completion 0.00001 per token."},{"capture_id":"fd80ded2e526ed31478cc9c61df580c27ced3bf2cf162539fb257bd2e7f15a8f","disposition":"included","statement":"The 20 October capture of https://claude.com/pricing lists Max 5x and Max 20x. Its visible text does not contain Claude Duet or Ultra. Opus 5.5 is input $4 and output $20 per million tokens. Sonnet 5.5 is input $2 and output $10, with prompt-cache read $0.10."},{"capture_id":"8ed6957ba7cc0775dcaf10806865ba6e805a4d61930b6173d2612ac6b6a75dd2","disposition":"included","statement":"The 21 October capture of https://claude.com/pricing contains the navigation label Claude Duet in the header and footer. No other visible line names Duet. The page lists Max 5x and Max 20x and does not contain Ultra. Opus 5.5 is input $4 and output $20 per million tokens. Sonnet 5.5 is input $2 and output $10, with prompt-cache read $0.10."},{"capture_id":"4c3605a4ec142c2d51984278c2cfc49837398ef6f9bc2e8f6dbd922411a87182","disposition":"included","statement":"The 25 October capture of https://claude.com/pricing lists Opus 5.5 at input $4 and output $20 per million tokens, with prompt-cache read $0.20 and cache write $5. Sonnet 5.5 is input $2 and output $10, with prompt-cache read $0.10 and cache write $2.50. Haiku 5.5 is input $0.10 and output $0.50 per million tokens for prompts of at most 100,000 tokens, and input $0.50 and output $2.50 for longer prompts, with cache read $0.01 and $0.05 and cache write $0.125 and $0.625 on those two tiers. The page lists Max 5x and Max 20x, contains the navigation label Claude Duet, and does not contain Ultra."},{"disposition":"excluded","exclusion_reason":"The OpenRouter row is a separate included claim. It does not establish a percentage cut, a cost per task, a parameter count, a weights date, or an index score. No Moonshot, Artificial Analysis, or Epoch capture is in the ledger.","statement":"Kimi K3.1 Pro's Highspeed variant, the promised full weights, the 33% price cut, the cost of about $2.11 per task, the 2.8T/104B mixture-of-experts description, and the Artificial Analysis rank, Elo, and Terminal-Bench scores."},{"disposition":"excluded","exclusion_reason":"The pricing page adds a navigation label and does not describe the product. No Anthropic Duet announcement is captured.","statement":"Claude Duet as a research preview on a waitlist for US Pro and Max, a proactive Opus 5.5 agent with connected apps, its own cloud VM, and approval for consequential actions."},{"disposition":"excluded","exclusion_reason":"None of those primaries are captured in the ledger.","statement":"OpenAI's end of live-internet access for frontier-model training, the PHP disclosures, the OpenAI financing reports, the AWS capacity statement, the UK announcement, the EU harmonized standard, Anthropic's draft bill, and Moonshot's statement about training material together with China's draft export controls."},{"disposition":"excluded","exclusion_reason":"No Artificial Analysis or Epoch capture is in the ledger. The named Claude rates are a separate included claim.","statement":"The brief's Frontier cluster ordering, Epoch Capabilities Index figures, and the claim of no major broad price cuts."},{"disposition":"excluded","exclusion_reason":"Those labels need a task, five instruments, and comparable profiles. The compendium proposals, economics narrative, task-router notes, and look-ahead items depend on uncaptured announcements and index scores.","statement":"The brief's Frontier, Efficient, and Budget labels, its compendium accept, watch, and reject proposals, the economics narrative beyond the captured token fields, the task-router atlas notes, and the look-ahead items."}],"reviewed_at":"2026-10-31T17:22:56+01:00","reviewer":"grok","schema_version":1,"title":"Week 43 brief against ledger captures"}
|
||||
File diff suppressed because one or more lines are too long
|
|
@ -0,0 +1,33 @@
|
|||
# Reviewed publication
|
||||
|
||||
This file is a reviewed publication. It is not a generated daily, weekly, or external draft.
|
||||
|
||||
- Revision: `2181566caba12b0325c604b53168a6c02acaeddcc88cefaf3963d1a46ee60f23`
|
||||
- Title: Week 43 brief against ledger captures
|
||||
- Reviewer: grok
|
||||
- Reviewed: 2026-10-31T17:22:56+01:00
|
||||
|
||||
## Claims
|
||||
|
||||
- included: The 22 October OpenRouter models capture does not list moonshotai/kimi-k3.1-pro.
|
||||
- Capture: `0dfc47ac722749067c30845c598b72023ebbabc0793b386bdac961bd4e65fa28`
|
||||
- included: The 23 October OpenRouter models capture lists moonshotai/kimi-k3.1-pro as MoonshotAI: Kimi K3.1 Pro, context 1048576, prompt 0.000002 and completion 0.00001 per token. The row does not state a percentage price cut, a parameter count, a weights-release date, or an index score.
|
||||
- Capture: `252f74fe0aa7a965a8a875817768a9c1bdda4b9b765889c985080e230230c8cd`
|
||||
- included: The 25 October OpenRouter models capture lists moonshotai/kimi-k3.1-pro as MoonshotAI: Kimi K3.1 Pro, context 1048576, prompt 0.000002 and completion 0.00001 per token.
|
||||
- Capture: `97562814bb3287f7d90aac8e61bc8ae1d3e72c29d5c41438c65a22ca8c9c56c0`
|
||||
- included: The 20 October capture of https://claude.com/pricing lists Max 5x and Max 20x. Its visible text does not contain Claude Duet or Ultra. Opus 5.5 is input $4 and output $20 per million tokens. Sonnet 5.5 is input $2 and output $10, with prompt-cache read $0.10.
|
||||
- Capture: `fd80ded2e526ed31478cc9c61df580c27ced3bf2cf162539fb257bd2e7f15a8f`
|
||||
- included: The 21 October capture of https://claude.com/pricing contains the navigation label Claude Duet in the header and footer. No other visible line names Duet. The page lists Max 5x and Max 20x and does not contain Ultra. Opus 5.5 is input $4 and output $20 per million tokens. Sonnet 5.5 is input $2 and output $10, with prompt-cache read $0.10.
|
||||
- Capture: `8ed6957ba7cc0775dcaf10806865ba6e805a4d61930b6173d2612ac6b6a75dd2`
|
||||
- included: The 25 October capture of https://claude.com/pricing lists Opus 5.5 at input $4 and output $20 per million tokens, with prompt-cache read $0.20 and cache write $5. Sonnet 5.5 is input $2 and output $10, with prompt-cache read $0.10 and cache write $2.50. Haiku 5.5 is input $0.10 and output $0.50 per million tokens for prompts of at most 100,000 tokens, and input $0.50 and output $2.50 for longer prompts, with cache read $0.01 and $0.05 and cache write $0.125 and $0.625 on those two tiers. The page lists Max 5x and Max 20x, contains the navigation label Claude Duet, and does not contain Ultra.
|
||||
- Capture: `4c3605a4ec142c2d51984278c2cfc49837398ef6f9bc2e8f6dbd922411a87182`
|
||||
- excluded: Kimi K3.1 Pro's Highspeed variant, the promised full weights, the 33% price cut, the cost of about $2.11 per task, the 2.8T/104B mixture-of-experts description, and the Artificial Analysis rank, Elo, and Terminal-Bench scores.
|
||||
- Excluded because: The OpenRouter row is a separate included claim. It does not establish a percentage cut, a cost per task, a parameter count, a weights date, or an index score. No Moonshot, Artificial Analysis, or Epoch capture is in the ledger.
|
||||
- excluded: Claude Duet as a research preview on a waitlist for US Pro and Max, a proactive Opus 5.5 agent with connected apps, its own cloud VM, and approval for consequential actions.
|
||||
- Excluded because: The pricing page adds a navigation label and does not describe the product. No Anthropic Duet announcement is captured.
|
||||
- excluded: OpenAI's end of live-internet access for frontier-model training, the PHP disclosures, the OpenAI financing reports, the AWS capacity statement, the UK announcement, the EU harmonized standard, Anthropic's draft bill, and Moonshot's statement about training material together with China's draft export controls.
|
||||
- Excluded because: None of those primaries are captured in the ledger.
|
||||
- excluded: The brief's Frontier cluster ordering, Epoch Capabilities Index figures, and the claim of no major broad price cuts.
|
||||
- Excluded because: No Artificial Analysis or Epoch capture is in the ledger. The named Claude rates are a separate included claim.
|
||||
- excluded: The brief's Frontier, Efficient, and Budget labels, its compendium accept, watch, and reject proposals, the economics narrative beyond the captured token fields, the task-router atlas notes, and the look-ahead items.
|
||||
- Excluded because: Those labels need a task, five instruments, and comparable profiles. The compendium proposals, economics narrative, task-router notes, and look-ahead items depend on uncaptured announcements and index scores.
|
||||
|
|
@ -0,0 +1,37 @@
|
|||
# Reviewed publication
|
||||
|
||||
This file is a reviewed publication. It is not a generated daily, weekly, or external draft.
|
||||
|
||||
- Revision: `43113feeff26834ad3d7585469141fe2c13a4652b7c650500d4e4e23afba1630`
|
||||
- Title: Week 44 daily briefs against ledger captures through 30 October
|
||||
- Reviewer: grok
|
||||
- Reviewed: 2026-10-31T17:28:42+01:00
|
||||
|
||||
## Claims
|
||||
|
||||
- included: The 25 October OpenRouter models capture lists neither openai/gpt-6.1-terra nor google/gemini-4-carbon, and no model id contains rosalind. It lists mistralai/mistral-large-4-0 at context 1048576, prompt 0.00000136 and completion 0.00000418 per token.
|
||||
- Capture: `97562814bb3287f7d90aac8e61bc8ae1d3e72c29d5c41438c65a22ca8c9c56c0`
|
||||
- included: The 27 October OpenRouter models capture lists openai/gpt-6.1-terra and openai/gpt-6.1-terra-pro as GPT-6.1 Terra and GPT-6.1 Terra Pro, context 1050000, prompt 0.000001, completion 0.000005, and input-cache read 0.00000005 per token. The batch ids openai/gpt-6.1-terra:batch and openai/gpt-6.1-terra-pro:batch are prompt 0.0000005 and completion 0.0000025. On the same capture, openai/gpt-5.6-terra is prompt 0.000002 and completion 0.000012, and openai/gpt-6.1-sol is prompt 0.000002 and completion 0.00001. The capture does not list google/gemini-4-carbon or an id containing rosalind, and it does not state a benchmark score or a ChatGPT surface.
|
||||
- Capture: `f5c006d2c66151f42867d5825485ce89c01acda9ef6bcea273b2486187d385ec`
|
||||
- included: The 28 October OpenRouter models capture lists mistralai/mistral-large-4-0 at context 1048576, prompt 0.00000136 and completion 0.00000418 per token. It does not list google/gemini-4-carbon.
|
||||
- Capture: `f448f544642b9b130744ce4c672f5dcda8e96de98a8595b1402e59e2fd6baf8d`
|
||||
- included: The 29 October OpenRouter models capture lists mistralai/mistral-large-4-0 as Mistral: Mistral Large 4, context 1048576, prompt 0.0000007 and completion 0.0000021 per token. The row does not state a parameter count, a license, or a benchmark score.
|
||||
- Capture: `f445ff144b564ca63cbce30a8e871cbf1e226e4f13d742f2ca50da40542585c1`
|
||||
- included: The 30 October OpenRouter models capture lists openai/gpt-6.1-terra at context 1050000, prompt 0.000001, completion 0.000005, and input-cache read 0.00000005 per token, and mistralai/mistral-large-4-0 at context 1048576, prompt 0.0000007 and completion 0.0000021 per token. It does not list google/gemini-4-carbon or an id containing rosalind.
|
||||
- Capture: `77d301db08d22a6a15408cdc2aeb71f9ed4855804bd6b98003d2747f6e2dc168`
|
||||
- included: The 29 October capture of https://claude.com/pricing lists the plan names Max 5x and Max 20x and the words From $100. It does not contain Ultra. Opus 5.5 is input $4 and output $20 per million tokens. Sonnet 5.5 is input $2 and output $10, with prompt-cache read $0.10.
|
||||
- Capture: `4ab8811e9a4d0086b5770fff74fcfd3774ddd0c8f5593b47bb05ec1f38dd9db4`
|
||||
- included: The 30 October capture of https://claude.com/pricing lists Max at $100 a month and Ultra at $200 a month. Max lists 5x Pro usage every 5 hours and every week, Fable included up to half the weekly limit, and $100 a month in platform API credits. Ultra lists no 5-hour limit for verified accounts, 10x Pro usage every week, and $200 a month in platform API credits, up from $100 on Max. Pro lists Fable paid with usage credits. The visible text does not contain the plan names Max 5x or Max 20x. Opus 5.5 is input $4 and output $20 per million tokens, with prompt-cache read $0.20 and cache write $5. Sonnet 5.5 is input $2 and output $10, with prompt-cache read $0.10 and cache write $2.50. Haiku 5.5 is input $0.10 and output $0.50 per million tokens for prompts of at most 100,000 tokens, and input $0.50 and output $2.50 for longer prompts, with cache read $0.01 and $0.05 and cache write $0.125 and $0.625 on those two tiers.
|
||||
- Capture: `83ee9a876a57f393662514e3cf4f09535fedfcde01856c451fd1cb3fd7177722`
|
||||
- excluded: GPT-6.1 Terra's benchmark scores, speed claims, ChatGPT availability, and the statement that its price is 50% below the prior Terra rate card.
|
||||
- Excluded because: The OpenAI announcement is not captured. The OpenRouter fields are a separate included claim. On that 27 October capture the GPT-6.1 Terra completion field is 0.000005 and the GPT-5.6 Terra completion field is 0.000012.
|
||||
- excluded: Gemini 4 Carbon's announcement, the Artificial Analysis score of 59, the cyber-benchmark percentages, Fairwind access, and introductory pricing of $2 and $10 followed by $4 and $20.
|
||||
- Excluded because: The Google blog and the Artificial Analysis posts are not captured. The 30 October OpenRouter capture does not list a carbon id, and that absence does not establish an access program or a later price.
|
||||
- excluded: Mistral Large 4 open weights, the parameter count, the license terms, and the Terminal-Bench and FrontierCode scores.
|
||||
- Excluded because: The Mistral announcement is not captured. The OpenRouter price fields are a separate included claim and do not establish a parameter count, a license, or a benchmark score.
|
||||
- excluded: GPT-6 Rosalind Research, Genie 4, the Anthropic S-1 and its revenue figures, the compute-financing announcements, the EU and UK items, the GitHub figures, and the Claude Duet product description.
|
||||
- Excluded because: None of those primaries are captured. The 30 October OpenRouter capture does not list an id containing rosalind, and that absence does not describe the product.
|
||||
- excluded: Artificial Analysis Index scores and version changes, and the Epoch figures.
|
||||
- Excluded because: No Artificial Analysis or Epoch capture is in the ledger.
|
||||
- excluded: The briefs' Frontier, Efficient, and Budget labels and their ledger-action proposals.
|
||||
- Excluded because: Those labels need a task, five instruments, and comparable profiles. The ledger-action proposals depend on uncaptured announcements and index scores.
|
||||
|
|
@ -240,17 +240,12 @@ Source content was captured or changed. Extracted claims require editorial revie
|
|||
|
||||
## Reviewed reading
|
||||
|
||||
Week 43 is Monday 19 October through Sunday 25 October. The Anthropic pricing page still showed Max 5x and Max 20x on every capture in this week. The Ultra wording begins on 30 October and is filed in week 44.
|
||||
The Grok week-43 brief is retained at `digests/external/weekly/2026-W43-grok.md` and stays unverified as a whole. The pass against ledger captures is publication `2181566caba12b0325c604b53168a6c02acaeddcc88cefaf3963d1a46ee60f23`.
|
||||
|
||||
OpenRouter membership that first appeared this week:
|
||||
Included from captures: the 22 October OpenRouter capture does not list `moonshotai/kimi-k3.1-pro`. The 23 October capture lists it as MoonshotAI: Kimi K3.1 Pro, context 1048576, prompt `0.000002` and completion `0.00001` per token. The 25 October capture lists those same fields. The 20 October pricing capture lists Max 5x and Max 20x, Opus 5.5 at $4 / $20 per million tokens, and Sonnet 5.5 at $2 / $10 with cache read $0.10. It does not contain Claude Duet or Ultra. The 21 October pricing capture adds the navigation label Claude Duet and no other visible line names Duet. The 25 October pricing capture lists the same Opus and Sonnet rates, the Haiku 5.5 tiers at $0.10 / $0.50 and $0.50 / $2.50, Max 5x and Max 20x, and the Duet label.
|
||||
|
||||
- 22 October: MiniMax M3.1 Flash and Recursive Flint (free) added; the M3.1 Flash free id and Gemini 2.5 Flash Lite batch id removed. `9c655f2f661980568a1386f4683e62e98e9c8b141a797377c27c7db11dee845e`
|
||||
- 23 October: Kimi K3.1 Pro added. `c4780101d764e78289c98b74b92423bc0434e23244097c6fead5d00ca3e508bb`
|
||||
- 24 October: three GPT-3.5 Turbo ids removed. `d56c7f465ac750d6462597528e776aa6dce82fe15edb561504acf7dbbad78d4f`
|
||||
- 25 October: LFM2.5-2.6B added. `30bf07927ac84e23aeec281c2807b8060d95c7868c86bba1bbe0b82f50de1ffe`
|
||||
|
||||
Prompt and completion fields moved by large factors on 21–25 October, and later captures reversed many of those moves. No stable price was filed inside this week. The 20-to-30 October endpoint comparison is `f89f102c67d1ad25ccc8e54e9f7bfd0674fefb37d6d5d602b034f170e1318b82`.
|
||||
Left out: the Kimi weights promise, the 33% cut, the cost-per-task figure, the parameter count, and the index scores; the Duet product description; the training-isolation announcement; the PHP disclosures; the financing, capacity, UK, EU, and draft-bill items; and Moonshot's training-material statement. Frontier, Efficient, and Budget are not assigned. Price readings are in `frontier/readings-2026-W43.md`.
|
||||
|
||||
## Editorial follow-up
|
||||
|
||||
No compendium edit. No frontier-ledger patch. Source-update sections above remain unextracted change notices. 25 October had captures and, before this pass, no daily draft; that draft now exists.
|
||||
Source-update sections above remain unextracted change notices. This pass does not make the external brief a belief.
|
||||
|
|
|
|||
|
|
@ -240,17 +240,12 @@ The 30 October capture of Anthropic's pricing page presents Max at $100 a month
|
|||
|
||||
## Reviewed reading
|
||||
|
||||
Week 44 runs Monday 26 October through Sunday 1 November. This draft stops at the 30 October captures. 31 October's 21:30 Europe/Berlin job had not run when these claims were written.
|
||||
Week 44 runs Monday 26 October through Sunday 1 November. No Grok weekly file for this week is retained. This pass reads the daily briefs for 26–30 October. There is no 26 October capture and no 31 October capture. The pass against ledger captures is publication `43113feeff26834ad3d7585469141fe2c13a4652b7c650500d4e4e23afba1630`.
|
||||
|
||||
Decision-relevant readings already listed above:
|
||||
Included from captures: the 25 October OpenRouter capture does not list `openai/gpt-6.1-terra`. The 27 October capture lists `openai/gpt-6.1-terra` and `openai/gpt-6.1-terra-pro` at context 1050000, prompt `0.000001`, completion `0.000005`, and input-cache read `0.00000005` per token. The batch ids are half those prompt and completion fields. On that capture, `openai/gpt-5.6-terra` is prompt `0.000002` and completion `0.000012`, and `openai/gpt-6.1-sol` is prompt `0.000002` and completion `0.00001`. The 30 October capture still lists the GPT-6.1 Terra fields. `mistralai/mistral-large-4-0` is prompt `0.00000136` and completion `0.00000418` on 28 October, and prompt `0.0000007` and completion `0.0000021` on 29 and 30 October. The 29 October pricing capture lists Max 5x and Max 20x and does not contain Ultra. The 30 October pricing capture lists Max at $100 a month and Ultra at $200 a month, with no 5-hour limit for verified accounts, 10x Pro usage every week, and $200 a month in platform API credits. Opus 5.5, Sonnet 5.5, and Haiku 5.5 token rows on that page are the rates already recorded for 25 October. The Sonnet page footer mission sentence changed. That sentence is outside the briefs.
|
||||
|
||||
- 27 October: GPT-6.1 Terra and Terra Pro, with batch ids, plus eight dropped ids (`dba698ff3377e1cc975350882a30ebccae7d8654f6895dabbad08391d61b522e`); Ternary Bonsai 2 9B the same evening (`76577fc1d15cc98ec285f46b9d987ee0867c15e9b683b9d1c982fcefdc33d74e`).
|
||||
- 28 October: Particle-1 added; GPT-4 Turbo, o1, o1-pro, o3-mini, o4-mini, and one Qwen thinking id removed (`96eb57948f37010512cd5fb3a0faed2ebab3b58c00f8585c2eb03be00e35b835`). Particle-1's per-token fields were lower again on 30 October.
|
||||
- 29 October: Solar Pro 3 removed (`b5d13a9bdea81682f00a4634786f7c2081a35fe72c29466f6c6d27a1d0a1b8c6`).
|
||||
- 30 October: Anthropic pricing page introduces Ultra at $200 a month beside Max at $100 (`fa7e93faf943cb94b02983b43b6aa2d7d81b5a7c10ed05576f4efe6e4cec243e`). OpenRouter endpoint price comparison against 20 October (`f89f102c67d1ad25ccc8e54e9f7bfd0674fefb37d6d5d602b034f170e1318b82`).
|
||||
|
||||
The Sonnet page and the Sonnet 5 launch note, compared with 20 October, changed the footer mission sentence. No product claim was extracted.
|
||||
Left out: Terra's benchmark scores, speed claims, and the blanket 50% cut against the prior Terra completion field; Carbon's announcement, index score, cyber-benchmark percentages, and later price; Mistral's weights, parameter count, license, and benchmark scores; Rosalind, Genie, the S-1, the financing items, and the Duet product description. Frontier, Efficient, and Budget are not assigned. Price readings are in `frontier/readings-2026-W44.md`.
|
||||
|
||||
## Editorial follow-up
|
||||
|
||||
Do not patch `frontier/ledger-2026-09.md` from this pass. Ultra is a plan on the pricing page. GPT-6.1 Terra is an OpenRouter id. The per-token moves are an endpoint snapshot with reversals in between. Source-update sections remain unextracted change notices.
|
||||
Source-update sections above remain unextracted change notices. This pass does not make the external briefs a belief. `frontier/ledger-2026-09.md` stays the September import.
|
||||
|
|
|
|||
7
frontier/readings-2026-W43.md
Normal file
7
frontier/readings-2026-W43.md
Normal file
|
|
@ -0,0 +1,7 @@
|
|||
# Price readings — 2026-W43
|
||||
|
||||
Read on 31 October 2026 from captures already in the ledger. These lines do not assign Frontier, Efficient, or Budget. They do not replace [ledger-2026-09.md](ledger-2026-09.md) or the week-40, week-41, and week-42 readings.
|
||||
|
||||
On 23 October, capture `252f74fe0aa7a965a8a875817768a9c1bdda4b9b765889c985080e230230c8cd` lists `moonshotai/kimi-k3.1-pro` as MoonshotAI: Kimi K3.1 Pro, context 1048576, prompt `0.000002` and completion `0.00001` per token. That is $2 and $10 per million tokens as a catalog field, not a vendor card and not a measured percentage cut. The 22 October capture `0dfc47ac722749067c30845c598b72023ebbabc0793b386bdac961bd4e65fa28` does not list that id. The 25 October capture `97562814bb3287f7d90aac8e61bc8ae1d3e72c29d5c41438c65a22ca8c9c56c0` lists the same prompt and completion fields.
|
||||
|
||||
On 25 October, capture `4c3605a4ec142c2d51984278c2cfc49837398ef6f9bc2e8f6dbd922411a87182` of https://claude.com/pricing lists Opus 5.5 at input $4 and output $20 per million tokens, Sonnet 5.5 at input $2 and output $10 with prompt-cache read $0.10, and the Haiku 5.5 tiers already recorded for 18 October in [readings-2026-W42.md](readings-2026-W42.md). The page lists Max 5x and Max 20x. It does not contain Ultra. The navigation label Claude Duet is present. The 20 October capture `fd80ded2e526ed31478cc9c61df580c27ced3bf2cf162539fb257bd2e7f15a8f` does not contain that label.
|
||||
9
frontier/readings-2026-W44.md
Normal file
9
frontier/readings-2026-W44.md
Normal file
|
|
@ -0,0 +1,9 @@
|
|||
# Price readings — 2026-W44
|
||||
|
||||
Read on 31 October 2026 from captures already in the ledger. These lines stop at 30 October. They do not assign Frontier, Efficient, or Budget. They do not replace [ledger-2026-09.md](ledger-2026-09.md) or the week-40 through week-43 readings.
|
||||
|
||||
On 27 October, capture `f5c006d2c66151f42867d5825485ce89c01acda9ef6bcea273b2486187d385ec` lists `openai/gpt-6.1-terra` and `openai/gpt-6.1-terra-pro` at context 1050000, prompt `0.000001`, completion `0.000005`, and input-cache read `0.00000005` per token. That is $1, $5, and $0.05 per million tokens as catalog fields. The batch ids are prompt `0.0000005` and completion `0.0000025`. On the same capture, `openai/gpt-5.6-terra` is prompt `0.000002` and completion `0.000012`, and `openai/gpt-6.1-sol` is prompt `0.000002` and completion `0.00001`. The 25 October capture `97562814bb3287f7d90aac8e61bc8ae1d3e72c29d5c41438c65a22ca8c9c56c0` does not list `openai/gpt-6.1-terra`. There is no 26 October capture. The 30 October capture `77d301db08d22a6a15408cdc2aeb71f9ed4855804bd6b98003d2747f6e2dc168` lists the same GPT-6.1 Terra prompt, completion, and cache-read fields.
|
||||
|
||||
On 29 October, capture `f445ff144b564ca63cbce30a8e871cbf1e226e4f13d742f2ca50da40542585c1` lists `mistralai/mistral-large-4-0` at context 1048576, prompt `0.0000007` and completion `0.0000021` per token. That is $0.70 and $2.10 per million tokens as catalog fields. The 28 October capture `f448f544642b9b130744ce4c672f5dcda8e96de98a8595b1402e59e2fd6baf8d` lists prompt `0.00000136` and completion `0.00000418`. The 30 October capture lists the 29 October fields.
|
||||
|
||||
On 30 October, capture `83ee9a876a57f393662514e3cf4f09535fedfcde01856c451fd1cb3fd7177722` of https://claude.com/pricing lists Max at $100 a month and Ultra at $200 a month. Ultra has no 5-hour limit for verified accounts, 10x Pro usage every week, and $200 a month in platform API credits. The Opus 5.5, Sonnet 5.5, and Haiku 5.5 token rows match the 25 October reading in [readings-2026-W43.md](readings-2026-W43.md). The 29 October capture `4ab8811e9a4d0086b5770fff74fcfd3774ddd0c8f5593b47bb05ec1f38dd9db4` lists Max 5x and Max 20x and does not contain Ultra.
|
||||
Loading…
Add table
Add a link
Reference in a new issue