Skip to content
GenAI BR
Português
MethodMethod v2.1 · October 2026

Why trust what the house publishes.

This is GenAI BR's charter: whom we write for, how an edition is made, what every number and every judgment means, and the scoreboard of what we got right and wrong. The page is versioned. When the method changes, it changes here before it changes in practice.

01

For whom

We write for the engineer who will put an agent into production on Tuesday, in Brazil.

She is a concrete person, not an audience. She has a system to ship, an inference budget, a committee asking "why not the open model?" and one hour a week to learn what actually changed. Cinco is written for that hour. If a paragraph does not change what she does on Tuesday, it goes.

Around her are two rings. Those who decide — purchases, hiring, architecture, regulation — read what she reads but need the result on one page: the second cut, Cinco for decision-makers, exists for that ring. And those starting out, who prefer five things explained to fifty headlines, and find the glossary and the trails in the Archive.

What the reader is not: an investor looking for a thesis, a journalist looking for a headline, a vendor looking for validation. They are welcome; the page is not written for them.

  1. 01
    The engineer

    Builds with AI and needs to know what actually changed, not what was announced. The whole essay is for her: the thesis, the movements, the code.

  2. 02
    Those who decide

    Buy, hire, design the architecture or regulate. They read the judgments and the panel; from October, they get the same week on one page.

  3. 03
    Those starting out

    Prefer five explained items to fifty headlines. They come in through the glossary and the Archive's trails, and climb when they want.

02

How an edition is made

One reporting, two cuts.

The week starts with a pile of a couple of hundred things — papers, repositories, official documents, announcements — and ends with a thesis carried by five movements. The funnel below shows the stages and the day of each. The initial triage is AI-assisted; the reading, the checking, the code and the text are one person's.

Two hundred become five

The funnel of a typical week. The numbers are averages of the first weeks and will be updated with each edition.

  1. ~200
    skimmed
    Monday morning

    Titles, abstracts and metadata of everything that came out: arXiv, repositories, announcements, official gazette. AI-assisted triage flags what deserves reading.

  2. ~40
    actually read
    Monday afternoon

    Full paper, open code, whole document. Here the AI leaves the room.

  3. 12
    verified
    Monday night

    Every number checked against the source. Self-reported marked. The arithmetic that carries the argument becomes code and is run.

  4. 5
    published
    Tuesday, 7am

    The movements that carry the week's thesis. Each primary source enters the Archive dated — and stays.

The filter

One question decides what gets in: does this change a builder's decision in the coming months? Five criteria hold it up.

  1. 1Primary source or nothing

    Every item leads to the paper, the code, the official document or the original announcement. Press coverage enters as complementary reading, never as the source.

  2. 2Result before opinion

    Each entry separates what was demonstrated from what was not. Self-reported numbers are marked as such.

  3. 3Relevance for builders

    What changes an engineering, product or purchasing decision in the coming months gets in. What is only noise stays out, even if it is news.

  4. 4Brazil when there is a document

    Regulation, public money and the labor market enter when there is an official text, a number or a deadline — not when there is only a promise.

  5. 5Conflicts declared

    When an item touches something the house has a relationship with, it is written in the entry.

The two cuts

It is The Economist's rule: one reporting, several forms. What changes between the cuts is density, never the conclusion.

  1. CincoLive since September 2026

    The essay. The thesis, five judgments with confidence levels, the panel, five movements with figures and executed code, what to watch, the numbered sources. Forty minutes of serious reading, every Tuesday at 7am.

  2. Cinco for decision-makersPlanned for October 2026

    The same week on one page: what changed, what to do with it, what is still unproven. Nothing in it that was not reported in the essay; no judgment differs between the two.

Roadmap→

Every edition has the same seven parts

On the left, the real edition no. 1. On the right, what each part promises. Repetition is what makes one week comparable to another — and accountable.

Cinco nº 7 · Tuesday, 20 October 2026
Verified: what changes when a third party measures
What each part promises
Opening

On 13 October, Gemini 4 Argon reached general availability and, the same day, the ARC Prize published its independent verification: 97.5% on ARC-AGI-3 with the harness designed with Google, 49.8% with the standard harness, the latter with runs that did not finish.[1,4] On 19 October, Vals AI published its Terminal-Bench 4.0 measurement, with Opus 5.5 ahead of Sonnet 5.5 — the opposite of what Anthropic says and of what Artificial Analysis shows.[6,7] Midweek, ANPD published the third report of its AI sandbox, with the first real tests.[17]

1
The thesis

Three paragraphs on what the week means, not what happened. Read only this and you leave knowing the house's position.

Judgments

1Through 31 March 2027, no model passes 70% on ARC-AGI-3 with the standard harness, in the ARC Prize's verification. · moderate confidence

2
Judgments

Five conclusions, each with a confidence level — high, moderate, low — in the vocabulary of intelligence analysis. The next edition revisits them.

Panel

100×The scissors — closed frontier ÷ permissive open

3
Panel

Six fixed indicators, read again every week: frontier price, open price, the scissors, two benchmark ceilings, public money that became a tender.

Movements

IIWho is better at the terminal depends on who measures · 2 figures, 1 code demos

4
Movements

Five chapters of the same argument. Each with a figure or an executed code demonstration — and the data table always one click away.

Method note

Self-reported on the model card: 65.7% on SWE-bench Pro. There is no technical report. None of the four scores has been replicated.

5
Method

Self-reported or independent, on every number. Sample, interval, harness. What we do not know is written down.

What to watch

31 Oct 2026, 16:00 UTCDeepSeek API search billing starts. By that date, too, the Mistral Large 4 weights promised in no. 6.

6
What to watch

Dates and conditions that would confirm or refute the judgments — so the next edition can be held to account.

Sources

1Gemini 4 Argon — model card · Google DeepMind · 20 sources

7
Sources

Numbered, dated, with what we took from each. Press only when it is the sole source, and said so.

03

What the numbers mean

A number in this house is a number someone published, with the source next to it. Never an estimate of ours presented as data.

Self-reported or independent
Every result says who measured it. A number from the lab itself is marked self-reported; a replication is called one. The two never share a bar without warning.
Bars always from zero
We never truncate the axis of a bar chart. If the difference only shows with a cut axis, the difference is not the argument.
One scale per chart, and the log declared
Every figure has one scale. When it is logarithmic — prices, parameters — it says so in the subtitle, and the axis ticks show it.
Direct labels, not a legend
The name sits next to the point or the line. A scatter without a name on every point is a cloud, not an argument.
Every figure has its table
Every chart carries its data as a table, one click away, with the source. What cannot be read as a table does not become a chart.
Executed code, not illustrative
When arithmetic carries the argument, it is written as code, it was run, and the published output is the real output, dated. If you run it and get something else, write in.
Sample and interval, when they exist
A 119-task benchmark does not separate two models by two points; we say when a difference is noise. When the interval was not published, we say it was not.
Numbers in Portuguese
US$ 0,50, 47,9%, 7.000 petaflops. Decimal comma and thousands point, as in the country where the page is read.

The judgment vocabulary

Every edition closes with judgments: what we conclude and how sure we are, in the vocabulary of intelligence analysis (the ICD 203 directive, adapted).

High confidence
Independent sources converge, the mechanism is known and the track record is good. We would be surprised to be wrong.
Moderate confidence
The evidence is plausible and consistent but has a gap: a single source, a self-reported number, a short horizon. It is our reading, and it may change.
Low confidence
The evidence is fragmentary or the track record is poor. We publish because the question matters, not because the answer is firm.

What happens next

Every judgment is born open and gets, in "What to watch", the dates and conditions that would confirm or refute it. The edition that passes those dates comes back to it — confirmed, refuted or revised — with a line on what happened. We never delete a wrong judgment; the scoreboard below is the sum of all of them.

open
Published, waiting for the dates in "What to watch".
confirmed
The condition happened. It stays on the scoreboard as a hit, with the edition that closed it.
refuted
The opposite happened, or the deadline passed without the condition. It stays on the scoreboard as a miss, with the same visibility.
revised
We changed the wording or the confidence before the deadline. The earlier text stays visible.
04

Indicators

Six fixed series, read again every edition. Each reading is a number someone published, with the source. When an indicator's method changes, the series restarts and the earlier one stays. The whole series is here and as CSV.

Download the series as CSV↓

  1. Closed frontier price

    7 readings
    US$ per million output tokens

    How it is measured. List price, short context, of the most expensive general-purpose model among OpenAI, Anthropic and Google on the edition's day.

    1. no. 1$5008 Sept 2026 · GPT-6 Astra and Claude Fable 5.1 · openai.com
    2. no. 2$5015 Sept 2026 · GPT-6 Astra and Claude Fable 5.1 · developers.openai.com
    3. no. 3$5022 Sept 2026 · GPT-6 Astra and Claude Fable 5.1 · developers.openai.com
    4. no. 4$5029 Sept 2026 · GPT-6 Astra and Claude Fable 5.1 · The model at the top of the Artificial Analysis index, Claude Opus 5.5, costs US$ 20: the most expensive is no longer the most capable. · developers.openai.com
    5. no. 5$5006 Oct 2026 · GPT-6 Astra and Claude Fable 5.1 · developers.openai.com
    6. no. 6$5013 Oct 2026 · GPT-6 Astra and Claude Fable 5.1 · developers.openai.com
    7. no. 7$5020 Oct 2026 · GPT-6 Astra and Claude Fable 5.1 · developers.openai.com

  2. Permissive open price

    7 readings
    US$ per million output tokens

    How it is measured. Since no. 2: list output price, on the lab's own API, of the open-weight model under MIT or Apache 2.0 ranked highest on the Artificial Analysis Intelligence Index. Up to no. 1 it was the lowest price among permissive open models with an index ≥ 55 — a cut that the index's new scale (v4.3, 7 September) left with no model at all.

    1. no. 1$0.5008 Sept 2026 · GLM-5.3-Flash (Z.ai) · huggingface.co
    2. series restarts. Artificial Analysis changed the index scale (v4.3) and no model, open or closed, reaches 55. The indicator now follows the highest-ranked permissive open model.no. 2$0.5015 Sept 2026 · GLM-5.3-Flash (Z.ai), index 42 · docs.z.ai
    3. no. 3$0.8722 Sept 2026 · MiMo-V2.6-Pro (Xiaomi), index 46 · artificialanalysis.ai
    4. no. 4$0.8729 Sept 2026 · MiMo-V2.6-Pro (Xiaomi), index 46 · artificialanalysis.ai
    5. no. 5$0.8706 Oct 2026 · MiMo-V2.6-Pro (Xiaomi), index 46 · mimo.mi.com
    6. no. 6$0.8713 Oct 2026 · MiMo-V2.6-Pro (Xiaomi), index 46 · mimo.mi.com
    7. no. 7$0.8720 Oct 2026 · MiMo-V2.6-Pro (Xiaomi), index 46 · mimo.mi.com

  3. The scissors

    7 readings
    closed frontier ÷ permissive open

    How it is measured. Ratio between the two indicators above. It measures the distance in price, not in capability.

    1. no. 1100×08 Sept 2026 · 50 ÷ 0.50 · Closed frontier price
    2. series restarts. Restarts with the permissive open price, whose method changed.no. 2100×15 Sept 2026 · 50 ÷ 0.50 · Permissive open price
    3. no. 358×22 Sept 2026 · 50 ÷ 0.87 · Permissive open price
    4. no. 458×29 Sept 2026 · 50 ÷ 0.87 · Permissive open price
    5. no. 558×06 Oct 2026 · 50 ÷ 0.87 · Permissive open price
    6. no. 658×13 Oct 2026 · 50 ÷ 0.87 · Permissive open price
    7. no. 758×20 Oct 2026 · 50 ÷ 0.87 · Permissive open price

  4. Terminal-Bench 4.0 ceiling

    7 readings
    best published score

    How it is measured. Highest published score on the current Terminal-Bench version, self-reported or independent — always stated which. When the version changes, the series restarts.

    1. no. 157.9%08 Sept 2026 · GPT-6 Astra, self-reported · openai.com
    2. no. 259.1%15 Sept 2026 · GPT-6 Astra, independent (Artificial Analysis) · Published on 7 September, the day no. 1 was closed; no. 1 read OpenAI's number. · artificialanalysis.ai
    3. no. 359.1%22 Sept 2026 · GPT-6 Astra, independent (Artificial Analysis) · artificialanalysis.ai
    4. no. 470.6%29 Sept 2026 · Claude Sonnet 5.5, self-reported · From the Sonnet 5.5 launch page, on 28 September. Six days earlier, Opus 5.5 had reported 66.4%. · anthropic.com
    5. no. 570.6%06 Oct 2026 · Claude Sonnet 5.5, self-reported · On the official leaderboard, which got Anthropic's entries the same day, Sonnet 5.5 shows 61.8% and Opus 5.5 64.85%. · anthropic.com
    6. no. 670.6%13 Oct 2026 · Claude Sonnet 5.5, self-reported · The Haiku 5.5 launch table, on 7 October, repeats the number. On the official leaderboard, Sonnet 5.5 has 61.8%. · anthropic.com
    7. no. 770.6%20 Oct 2026 · Claude Sonnet 5.5, self-reported · The week's highest independent number is 65.15%, Opus 5.5, at Vals AI. · anthropic.com

  5. SWE-bench Science ceiling

    7 readings
    bugs solved in scientific software

    How it is measured. Best model on the leaderboard kept by the OpenMOSS group (Fudan), independent of vendors. 119 tasks: differences of a few points are noise.

    1. no. 147.9%08 Sept 2026 · Claude Opus 5, max effort · swescience.github.io
    2. no. 250.4%15 Sept 2026 · GPT-6 Astra (max), with Codex · The row entered the leaderboard on 10 September, per the maintainers' repository history. · swescience.github.io
    3. no. 350.4%22 Sept 2026 · GPT-6 Astra (max), with Codex · swescience.github.io
    4. no. 450.4%29 Sept 2026 · GPT-6 Astra (max), with Codex · swescience.github.io
    5. no. 550.4%06 Oct 2026 · GPT-6 Astra (max), with Codex · swescience.github.io
    6. no. 650.4%13 Oct 2026 · GPT-6 Astra (max), with Codex · swescience.github.io
    7. no. 750.4%20 Oct 2026 · GPT-6 Astra (max), with Codex · swescience.github.io

  6. PBIA: money that became a tender

    7 readings
    R$ million, cumulative

    How it is measured. Sum of published tenders and calls with a fixed value and a deadline whose object declares Brazil's AI Plan (R$ 23 billion announced for 2024–2028). Announcements, partnerships without a tender and funds with a commitment range instead of a value do not count.

    1. no. 11,00908 Sept 2026 · Macaíba supercomputer (R$ 959M) and CNPq's RHAE IA call (R$ 50M) · Corrected: the draft of no. 1 said 959 and left out the RHAE IA call, published on 19 August. · facc.org.br
    2. no. 21,00915 Sept 2026 · Macaíba (R$ 959M) and RHAE IA (R$ 50M) · gov.br
    3. no. 31,00922 Sept 2026 · Macaíba and RHAE IA; no new tender · facc.org.br
    4. no. 41,00929 Sept 2026 · Macaíba and RHAE IA; no new tender · facc.org.br
    5. no. 51,00906 Oct 2026 · Macaíba and RHAE IA; no new tender · facc.org.br
    6. no. 61,00913 Oct 2026 · Macaíba (bids postponed to 4 Nov) and RHAE IA · facc.org.br
    7. no. 71,00920 Oct 2026 · Macaíba and RHAE IA; no new tender · facc.org.br
05

Judgment scoreboard

  • 35
    judgments
  • 35
    open
  • 0
    confirmed
  • 0
    refuted
  • 0
    revised

Every judgment published, from every edition, with its state. Generated from the editions' content, not typed by hand.

  1. no. 1 · 108 Sept 2026

    For 2026-class agentic work, cost per token no longer decides between closed and open. What decides is domain reliability, governance of the reasoning, and where the model runs.

    moderate confidenceopen
  2. no. 1 · 208 Sept 2026

    The ~100× gap between the closed frontier and permissive open models should persist for the next two quarters: closed labs are pricing reasoning and agency, not tokens, and open labs are pricing adoption.

    moderate confidenceopen
  3. no. 1 · 308 Sept 2026

    SWE-bench Verified no longer discriminates between frontier models; small-sample domain leaderboards do not separate the top five either. Whoever decides a purchase needs their own evaluation.

    high confidenceopen
  4. no. 1 · 408 Sept 2026

    Chain-of-thought monitorability will keep falling in flagships. “Reading the reasoning” should not be treated as a safety control in any production system.

    high confidenceopen
  5. no. 1 · 508 Sept 2026

    The Macaíba supercomputer will probably be contracted in 2026; it is unlikely to operate by the announced date (end of 2027), given three postponements and the absence of operating costs in the tender.

    low confidenceopen
  6. no. 2 · 115 Sept 2026

    By the end of 2026, another major API provider will do what DeepSeek did: serve an existing model name with a different model, at another price, by a changelog notice.

    moderate confidenceopen
  7. no. 2 · 215 Sept 2026

    The Agents API leaves beta by 31 March 2027, and the managed harness becomes OpenAI's default path for new agents.

    moderate confidenceopen
  8. no. 2 · 315 Sept 2026

    The next version of the Artificial Analysis index will move absolute scores enough to break any fixed cut. Whoever uses the index as a threshold needs a relative rule, like the one the panel adopted.

    high confidenceopen
  9. no. 2 · 415 Sept 2026

    GPT-6 Astra loses first place on SWE-bench Science by 31 March 2027.

    moderate confidenceopen
  10. no. 2 · 515 Sept 2026

    The RHAE IA call's result comes out by 31 December 2026 and commits the full R$ 50 million.

    low confidenceopen
  11. no. 3 · 122 Sept 2026

    Through 31 December 2026, the gap between the highest-ranked model on the Artificial Analysis index and the best open model under MIT or Apache 2.0 stays above five points.

    high confidenceopen
  12. no. 3 · 222 Sept 2026

    Through 31 March 2027, the highest-ranked permissive open model on the index keeps costing less than US$ 1.50 per million output tokens on the lab's own API.

    moderate confidenceopen
  13. no. 3 · 322 Sept 2026

    By 31 March 2027, Xiaomi publishes under MIT a MiMo scoring 50 or more on the Artificial Analysis index, in the version then current.

    low confidenceopen
  14. no. 3 · 422 Sept 2026

    First place on Artificial Analysis's voice index, held today by Grok Voice 3.0, changes hands at least once more by 31 March 2027.

    moderate confidenceopen
  15. no. 3 · 522 Sept 2026

    By 31 December 2026, xAI publishes how it ran the Terminal-Bench numbers in the Grok 4.7 launch table — harness and number of runs.

    low confidenceopen
  16. no. 4 · 129 Sept 2026

    Through 31 March 2027, the top-ranked model on the Artificial Analysis index does not again become the one with the highest output price among OpenAI, Anthropic and Google.

    moderate confidenceopen
  17. no. 4 · 229 Sept 2026

    Through 31 March 2027, neither Opus 5.5 nor Sonnet 5.5 reaches, on the official Terminal-Bench 4.0 leaderboard, the numbers Anthropic published for them (66.4% and 70.6%).

    high confidenceopen
  18. no. 4 · 329 Sept 2026

    By 31 March 2027, Anthropic allows switching off Opus 5.5's thinking through the API.

    low confidenceopen
  19. no. 4 · 429 Sept 2026

    By 31 March 2027, Artificial Analysis measures a model that spends more output tokens per task than Sonnet 5.5 at maximum effort.

    moderate confidenceopen
  20. no. 4 · 529 Sept 2026

    By 31 March 2027, Z.ai publishes under MIT or Apache 2.0 a model scoring above GLM-5.3 on the index.

    moderate confidenceopen
  21. no. 5 · 106 Oct 2026

    Gemini 4 Argon's price doubles on 1 December 2026, as announced: US$ 4 input and US$ 20 output.

    high confidenceopen
  22. no. 5 · 206 Oct 2026

    By 31 March 2027, OpenAI launches a successor to GPT-6 Astra at less than US$ 50 per million output tokens on the standard tier.

    moderate confidenceopen
  23. no. 5 · 306 Oct 2026

    By 31 March 2027, the official Terminal-Bench 4.0 leaderboard shows some model above 70%.

    moderate confidenceopen
  24. no. 5 · 406 Oct 2026

    The next model retirement Anthropic announces by 31 March 2027 gives less than 90 days between notice and shutdown.

    moderate confidenceopen
  25. no. 5 · 506 Oct 2026

    By 31 March 2027, Anthropic creates a speed tier above the current fast mode for some Claude model.

    low confidenceopen
  26. no. 6 · 113 Oct 2026

    By 31 March 2027, OpenAI or Anthropic launches a model with an output price below US$ 0.50 per million tokens.

    moderate confidenceopen
  27. no. 6 · 213 Oct 2026

    Mistral Large 4's weights come out by 30 November 2026, under Apache 2.0 or MIT.

    moderate confidenceopen
  28. no. 6 · 313 Oct 2026

    The Macaíba supercomputer session takes place on 4 November 2026, with no further postponement.

    moderate confidenceopen
  29. no. 6 · 413 Oct 2026

    The Macaíba supercomputer contract is signed by 31 December 2026.

    low confidenceopen
  30. no. 6 · 513 Oct 2026

    When Haiku 5.5 appears on the official Terminal-Bench 4.0 leaderboard, it will sit within five points of the 39.2% Anthropic published.

    moderate confidenceopen
  31. no. 7 · 120 Oct 2026

    Through 31 March 2027, no model passes 70% on ARC-AGI-3 with the standard harness, in the ARC Prize's verification.

    moderate confidenceopen
  32. no. 7 · 220 Oct 2026

    By 31 March 2027, DeepSeek again swaps the model that answers to the name deepseek-flash.

    high confidenceopen
  33. no. 7 · 320 Oct 2026

    By 31 December 2026, the license of DeepSeek V4.1-Flash-1017's weights shows up as MIT in the repository.

    moderate confidenceopen
  34. no. 7 · 420 Oct 2026

    By 31 March 2027, the rapporteur presents an opinion on PL 2338/2023 in the Chamber's special committee.

    low confidenceopen
  35. no. 7 · 520 Oct 2026

    By 31 March 2027, the ARC Prize publishes a verification of Gemini 4 Argon on the standard harness with all runs complete.

    low confidenceopen
06

What we do not do

We do not train models, sell open courses or run events
We read, check and write — and, with what we read, we build a product and advise companies. The money comes from the consulting (briefings, workshops, diagnostics and pilots) and will come from Spectrolite and, from 2027, from the Radar and State of AI Brasil. Reading Cinco is free.
We do not publish without a primary source
An item without a paper, code, document or original announcement does not get in, however newsworthy. Press is complementary reading, and said so.
We do not forecast without a confidence level
Every judgment says how sure we are and what would refute it. Opinion without that is not published as a judgment.
We do not delete what we got wrong
A refuted judgment stays on the scoreboard. A factual error is corrected in the entry with the earlier text visible, and listed in the corrections.
We do not take disguised paid content
Sponsorship of artifacts — Radar, State of AI — only without editorial influence and with the name on the cover. A paid item dressed as an entry, never.
We do not take vendor commissions
No AI vendor pays the house for referrals, resale or licenses — not in Cinco, not in the consulting.
We do not recommend Spectrolite in Cinco
It can be a subject, with the declaration at the top of the edition; never a recommendation, an offer or a success story. The product's results live on its own page.
We do not turn clients into stories
Nothing the consulting sees becomes Cinco. A case is only published with written permission, and with the relationship declared.
We do not let AI decide what matters
It triages and organizes; the choice, the checking and the text are one person's. Below it says exactly where it comes in.
We do not track who reads
No third-party scripts. The only form asks for an e-mail and says where it goes.
We do not publish more to look bigger
One edition a week, the same format. When the house grows, it will be to read deeper.
07

Declared interests

The house reads, builds and advises. Every relationship that could bend what it writes is here — and Cinco's checker enforces the declaration.

Who is on the register
Every commercial relationship that could influence what we publish: the house's product, its suppliers, whoever distributes it and the consulting clients. A new relationship enters the day there is a contract, a payment or a sale.
How it shows in an edition
When an edition names someone on the register, a line at the top, under the byline, says what the relationship is. The same line goes in the e-mail.
Who enforces it
Cinco's checker searches the edition's whole text — prose, figures, code, sources — for the names on the register and will not let the edition go out without the declaration. Spectrolite's competitors are not searched by name: that is the writer's judgment, and a missing declaration goes to the corrections.
What never happens
Cinco does not recommend Spectrolite or publish its results. No vendor pays the house for referrals. Nothing from the consulting becomes a story.

The register

WhoRelationshipSince
SpectroliteThe house's productGenAI BR builds, sells and answers for Spectrolite. Cinco may cover it as a subject; it never recommends it.Oct 2026
Google · GeminiSpectrolite supplierSpectrolite's model is Gemini 3.8 Flash. When an edition covers Google or Gemini, the declaration says so.Oct 2026
OpenRouterSpectrolite supplierIt is how Spectrolite reaches the model, with zero retention switched on in the account.Oct 2026
AppleSpectrolite distributorSpectrolite ships only through the App Store, with GenAI BR as the seller; Apple keeps a share of what is sold there.Oct 2026

Consulting clients: none so far. Each one enters here — by name, if they allow it, or by sector — the day the contract is signed.

The consulting's rules→Spectrolite→

08

AI policy

We use AI where it is good and say exactly where. If this changes, this page changes first.

What AI does here4
Triage
Filter hundreds of items a week, extract metadata (authors, date, source type, code availability) and flag what deserves reading.
Deduplication
Group the same result published in three different places.
Translation
The English version of this site is AI-translated from Portuguese and reviewed by a person. Portuguese is the original.
Code
This site was designed and built with an AI assistant, under human direction and review, decision by decision. It is described in "How this site was made", on the About page.
What AI never does4
Choose the movements
Selection is human. AI suggests reading; it does not decide what matters.
Write the final text
No published paragraph is machine-generated. What you read in an edition or an entry was written by the person who read the source.
Verify numbers
Every number is checked by a person against the primary source. AI gets numbers wrong with confidence; that is why it does not do this part.
Sign
Every edition carries a signature. It is a person, with a name, answering for what is written.
09

Corrections

Errors are corrected in the entry or the edition, with the date and the earlier text visible, and listed here. An error that changes a movement's conclusion gets a note in the following edition.

  1. 23 Oct 2026Cinco no. 1 — The draft's panel counted R$ 959 million on the PBIA indicator and left out CNPq's RHAE IA call (R$ 50 million, published on 19 August, aligned with the plan's Axis 4). The reading became 1,009.
  2. 23 Oct 2026Cinco no. 1 — A "What to watch" item waited for ARC Prize to verify GPT-6 Astra's ARC-AGI-3 result. The verification had come out on 2 and 3 September, before the edition was closed: 62.7% with the standard harness and 99.9% with the vendor's adapter.
10

Versions

Every change to the method gets a line here, dated. A change in practice without a change here is an error of ours — and belongs in the corrections.

  1. v2.123 Oct 2026

    An edition published after its Tuesday carries the actual publication date in its header; if it was written after the week it covers, its judgments only look past that date, and the checker enforces it. On the panel, the permissive open-price indicator swapped the fixed index cut (≥ 55), which Artificial Analysis's new scale left with no model, for the highest-ranked MIT or Apache 2.0 model; the series restarts at no. 2, and each reading may carry a note and a restart mark.

  2. v222 Oct 2026

    The house starts selling Spectrolite and consulting, and the register of interests comes in: every relationship that could influence what we publish is made public, and Cinco's checker enforces the declaration in the editions. "What we do not do" gains three lines — no vendor commissions, Spectrolite never recommended in Cinco, clients never become stories — and the money sentence changes. "We do not sell courses" becomes "we do not sell open courses": closed workshops for client teams are part of the consulting.

  3. v108 Sept 2026

    First version. Gathers what was spread across the About page — how we choose, the anatomy of an edition, the AI policy, the corrections — and adds the reader, the two cuts, the judgment vocabulary, the rules for numbers, the scoreboard and the indicator series.

Get the next Cinco

Tuesday, 7am BRT, in your inbox. Five items, with sources. No daily newsletter, no promotions, and unsubscribing is one click.

Just the e-mail. No third-party trackers on this page. Privacy policy.