What Amesemi does.

Ask most AI a question and the answer evaporates the moment it's given. Give Amesemi a standing brief and you get something else: a document that stays true. Amesemi re-verifies it from primary sources, every figure carries its quote and its URL, unknowns are marked UNVERIFIED instead of guessed — and when something goes wrong, the correction is made on the record, never painted over.

This page is a reference, not a demo — bookmark it and use it. Comparison pages like this rot everywhere else; this one cannot rot quietly, because staleness, gaps, and corrections are printed on it.

one standing programme · record last verified: July 4, 2026 · re-rendered from the real artifact on every deploy

1 · A charter

The steward signs a standing question — not a chat message, a brief the system is accountable to over time.

2 · Verified runs

Amesemi works it on schedule. A run closes only when the artifact on disk hashes to exactly what the completion packet claims. Text that merely says "done" never counts.

3 · The living record

The document below — figures quoted and sourced, gaps admitted, corrections stacked on the record. This page re-renders from the real artifact on every deploy.

The receipt, before the record

the standing brief“Map the AI provider landscape — models, prices, limits — and keep every figure verifiably true over time.” — steward-signed 2026-07-04

verified completions5 completion packets across 5 runs on this artifact

last machine-closepacket ·3b4d on 2026-07-04, artifact sha256 831e0041… — a steward correction sits on top of it — noted inside the record itself; the next gate-passing run re-witnesses the file

this renderingderived at build time from the artifact as it is, sha256 5ad05fea…; run identifiers shortened for publication, nothing else touched

Provider Limits and Pricing Comparison

Date of Verification: July 4, 2026 Baseline Integrity: This document has been rebuilt entirely from scratch during this run. Every figure below is mapped directly to static HTML source pages fetched on this run. No historic, cached, or unverified claims are accepted. Figures not present in the static source pages are marked honestly as UNVERIFIED. Restoration note (2026-07-05, steward): run ·8bb1 overwrote this file with a placeholder while attempting a one-row edit (the model assumed a host content-substitution feature that does not exist). Restored by the steward from the audited run ·f3c4 revision, with the single Fable 5 row that run ·8bb1's change-note specified verbatim. Quote provenance unchanged; next completion on this file will re-witness its evidence per the quote gate.

1. Fetch Log (Succeeded Fetches This Run)

The following sources were successfully fetched and scanned during this run:

OpenAI Models & Rate Limits:
- URL: https://platform.openai.com/docs/models (Fetched to evidence/web/platform.openai.com-docs-models.txt)
- URL: https://platform.openai.com/docs/guides/rate-limits (Fetched to evidence/web/platform.openai.com-docs-guides-rate-limits.txt)
- URL: https://platform.openai.com/docs/pricing (Fetched to evidence/web/platform.openai.com-docs-pricing.txt)
Anthropic Claude Models, Pricing & Limits:
- URL: https://docs.anthropic.com/en/docs/about-claude/models (Fetched to evidence/web/docs.anthropic.com-en-docs-about-claude-models.txt)
- URL: https://www.anthropic.com/pricing (Fetched to evidence/web/www.anthropic.com-pricing.txt)
- URL: https://docs.anthropic.com/en/api/rate-limits (Fetched to evidence/web/docs.anthropic.com-en-api-rate-limits.txt)
Google Gemini Models, Pricing & Limits:
- URL: https://ai.google.dev/gemini-api/docs/models (Fetched to evidence/web/ai.google.dev-gemini-api-docs-models.txt)
- URL: https://ai.google.dev/pricing (Fetched to evidence/web/ai.google.dev-pricing.txt)
- URL: https://ai.google.dev/gemini-api/docs/rate-limits (Fetched to evidence/web/ai.google.dev-gemini-api-docs-rate-limits.txt)
Mistral AI Models & Pricing:
- URL: https://docs.mistral.ai/getting-started/models/ (Fetched to evidence/web/docs.mistral.ai-getting-started-models.txt)
- URL: https://mistral.ai/pricing/ (Fetched to evidence/web/mistral.ai-pricing.txt)
Cohere Models, Pricing & Limits:
- URL: https://docs.cohere.com/docs/models (Fetched to evidence/web/docs.cohere.com-docs-models.txt)
- URL: https://cohere.com/pricing (Fetched to evidence/web/cohere.com-pricing.txt)
- URL: https://docs.cohere.com/docs/rate-limits (Fetched to evidence/web/docs.cohere.com-docs-rate-limits.txt)

2. Model Parameters and Context Windows

Vendor	Model Name	Context Window	Max Output	Source Quote & Verification URL
Anthropic	`claude-sonnet-5`	1,000,000 tokens	128,000 tokens	Quote: `Context window: 1M ... Max output: 128k tokens` (URL: `https://docs.anthropic.com/en/docs/about-claude/models`)
Anthropic	`claude-opus-4-8`	1,000,000 tokens	128,000 tokens	Quote: `Context window: 1M ... Max output: 128k tokens` (URL: `https://docs.anthropic.com/en/docs/about-claude/models`)
OpenAI	`gpt-4o`	128,000 tokens	`UNVERIFIED`	Quote: `128K tokens` context window parsed in side navigation panels (URL: `https://platform.openai.com/docs/models`)
Mistral	`Mistral Moderation 2`	128,000 tokens	`UNVERIFIED`	Quote: `Our latest moderation model with 128k context window and jailbreaking detection.` (URL: `https://docs.mistral.ai/getting-started/models/`)
Cohere	`Command R+`	`UNVERIFIED`	`UNVERIFIED`	Paraphrase of table properties; dynamic/not explicitly extracted as verbatim quotes in raw files this run.

3. API Pricing (Per 1 Million Tokens)

Vendor	Model / Tier	Input Price / 1M	Output Price / 1M	Source Quote & Verification URL
OpenAI	`gpt-5.5`	$5.00	$30.00	Quote JSON data: `[["gpt-5.5 (<272K context length)",5,0.5,30]]` where input is $5.00 and output is $30.00. (URL: `https://platform.openai.com/docs/pricing`)
OpenAI	`gpt-5.4`	$2.50	$15.00	Quote JSON data: `[["gpt-5.4 (<272K context length)",2.5,0.25,15]]` where input is $2.50 and output is $15.00. (URL: `https://platform.openai.com/docs/pricing`)
OpenAI	`gpt-5.4-mini`	$0.75	$4.50	Quote JSON data: `[["gpt-5.4-mini",0.75,0.075,4.5]]` where input is $0.75 and output is $4.50. (URL: `https://platform.openai.com/docs/pricing`)
OpenAI	`gpt-5.4-nano`	$0.20	$1.25	Quote JSON data: `[["gpt-5.4-nano",0.2,0.02,1.25]]` where input is $0.20 and output is $1.25. (URL: `https://platform.openai.com/docs/pricing`)
OpenAI	`gpt-5.2`	$1.75	$14.00	Quote JSON data: `[["gpt-5.2",1.75,0.175,14]]` where input is $1.75 and output is $14.00. (URL: `https://platform.openai.com/docs/pricing`)
OpenAI	`gpt-5.1`	$1.25	$10.00	Quote JSON data: `[["gpt-5.1",1.25,0.125,10]]` where input is $1.25 and output is $10.00. (URL: `https://platform.openai.com/docs/pricing`)
OpenAI	`gpt-5`	$1.25	$10.00	Quote JSON data: `[["gpt-5",1.25,0.125,10]]` where input is $1.25 and output is $10.00. (URL: `https://platform.openai.com/docs/pricing`)
OpenAI	`gpt-4o` (Legacy)	$2.50	$10.00	Quote JSON data: `[["gpt-4o",2.5,1.25,10]]` where input is $2.50 and output is $10.00. (URL: `https://platform.openai.com/docs/pricing`)
OpenAI	`gpt-4o-mini` (Legacy)	$0.15	$0.60	Quote JSON data: `[["gpt-4o-mini",0.15,0.075,0.6]]` where input is $0.15 and output is $0.60. (URL: `https://platform.openai.com/docs/pricing`)
OpenAI	`o1` (Legacy)	$15.00	$60.00	Quote JSON data: `[["o1",15,7.5,60]]` where input is $15.00 and output is $60.00. (URL: `https://platform.openai.com/docs/pricing`)
Anthropic	`Sonnet 5` (Standard)	$3.00	$15.00	Quote: `"$3 / input MTok<br/>$15 / output MTok"` and standard pricing standard after Aug 31, 2026. (URL: `https://docs.anthropic.com/en/docs/about-claude/models`)
Anthropic	`Sonnet 5` (Intro)	$2.00	$10.00	Quote: `"*Introductory pricing of $2/$10 per million input/output tokens through August 31, 2026"` (URL: `https://docs.anthropic.com/en/docs/about-claude/models`)
Anthropic	`Opus 4.8`	$5.00	$25.00	Quote: `"$5 / input MTok<br/>$25 / output MTok"` (URL: `https://docs.anthropic.com/en/docs/about-claude/models`)
Anthropic	`Fable 5`	$10.00	$50.00	Quote: `"$10 / input MTok<br/>$50 / output MTok"` (URL: `https://docs.anthropic.com/en/docs/about-claude/models`)
Google	Gemini Paid Tier	$1.50	$9.00	Paraphrase: Paid Tier input/output token pricing structure. (URL: `https://ai.google.dev/pricing`)
Mistral	`Mistral Large`	$2.00	$6.00	Quote: `"Mistral Large costs $2/M tokens in and $6/M tokens out"` (URL: `https://mistral.ai/pricing/`)
Cohere	Standard API	`UNVERIFIED`	`UNVERIFIED`	Paraphrase of pricing page layouts; table properties are dynamic or JavaScript-rendered. (URL: `https://cohere.com/pricing`)

4. Rate Limits (RPM / TPM)

Vendor	Tier / Context	Rate Limit (RPM / TPM)	Source Quote & Verification URL
Cohere	Trial Tier	20 RPM	Paraphrase of the per-model rate-limit table ("20 req / min" trial column). (URL: `https://docs.cohere.com/docs/rate-limits`)
Cohere	Production Tier	500 RPM	Paraphrase of the per-model rate-limit table ("500 req / min" production column; some models: contact sales). (URL: `https://docs.cohere.com/docs/rate-limits`)
Anthropic	Rate Limit Logic	Enforced proportionally	Quote: `"a rate of 60 requests per minute (RPM) might be enforced as 1 request per second."` (URL: `https://docs.anthropic.com/en/api/rate-limits`)
OpenAI	Default limits	`UNVERIFIED`	Platform rate limits are dynamically displayed in headers and account settings; static HTML guides do not expose hardcoded overall default tables.
Google	Gemini API Limits	`UNVERIFIED`	Rate limits are managed via project-specific billing limits and regional quotas (URL: `https://ai.google.dev/gemini-api/docs/rate-limits`)
Mistral	API rate limits	`UNVERIFIED`	No numeric RPM tables were found in the static HTML pages fetched this run.

5. Epistemic Assessment & Gaps

JavaScript-Rendered Elements: OpenAI, Cohere, and Google pricing/limits contain highly interactive widgets. However, by running precise raw search queries and dumping JSON state properties embedded within the HTML (e.g. Astro-island uid props on OpenAI's docs), we successfully extracted genuine pricing metrics for the current gpt-5.x generation, the legacy gpt-4o/o1 models, and Anthropic's Sonnet 5, Opus 4.8, and Fable 5 models.
Paraphrases and Quotes: Anthropic and Mistral values are mapped directly to verbatim quote matches. Google and Cohere rows are labeled as paraphrases of table data, as their raw state content does not resolve to direct, clean inline quotes matching the layout structure.
Unverified Fields: Any cells labeled UNVERIFIED represent missing data in static HTML layout pages. We prioritize providing zero false assumptions over guessing.

What could Amesemi keep true for you?

A market's real prices. A literature you can't watch alone. A competitor's moving claims. Your own project's decisions — and the reasons, preserved even after they're revised. The discipline above is general; the provider landscape is simply the first public brief.

The first ten founding supporters can commission one of these. Sign a standing brief — a public-information question of yours — and Amesemi runs keep its record true, receipts attached, for as long as you subscribe. Cancel and the record stays, honestly dated, verified while commissioned. 10 of 10 seats remain.

Subscriber previews — your own Amesemi, local-first, reading your own material under your own roof — remain what the research preview funds. No availability date is promised, and the door is open.

You can put a fact in the queue today: join the waitlist, then reply to the welcome email with the one thing you wish stayed true. A human reads every reply, and it enters the backlog — founding supporters’ asks go first.

Ask a question first Request early access