Vol. I · MMXXVI40+ models · 12 laboratories

Every frontier model,thinking on one page.

Abstract. Thought Field gathers the latest frontier models into a single, carefully typeset room for thinking — with reasoning you can read, sources you can check, and a memory that carries your context from one conversation to the next.

Keywords: model indexvisible reasoningpersonal memoryfiles & chatstypesettingplans

Begin — it's free Open a live demo Sign in

Thought Field

Anthropic/Claude Opus 4.7

Now asking

You

What did Wittgenstein mean by “the limits of my language are the limits of my world”?

Claude Opus 4.7

They want the idea behind the Tractatus aphorism, not jargon. Tie together language, thought, and the bounds of the expressible, then note the turn his later work takes.

For the early Wittgenstein, language pictures the world: whatever cannot be cast as a proposition falls outside the world we can think or share[1]. That boundary is less a wall than a horizon — ethics, the self, and the mystical sit at its edge, shown rather than said[2], and his later work dissolves the picture entirely, relocating meaning in use, in the language-games of a form of life.

[1]Tractatus Logico-Philosophicusgutenberg.org
[2]Philosophical Investigationsarchive.org

Fig. 1 — A live conversation: reasoning trace, streamed answer, and cited sources.

IThe catalogue

Every laboratory, one subscription.

Thought Field connects to each major lab through its official API, switching to a backup provider the moment one falters. No keys to manage, no tabs to juggle — open the index and choose.

Or let the field choose.

Undecided? Auto-routing reads your question and sends it to a fitting model — flagship depth for hard problems, speed for the rest.

OpenAI10 models
Anthropic4 models
Google4 models
DeepSeek4 models
Z.AI2 models
Minimax2 models

^° free tier^† reasoning^‡ web search

38 models

IIReasoning & sources

Show the work. Cite the sources.

Reasoning models think before they answer, and Thought Field keeps that thinking in a quiet fold above each reply — there when you want it, out of the way when you don't. When the web is consulted, every claim carries a numbered citation that resolves to a real reference beneath the text.

Citations are footnotes, not afterthoughts.

Thinking you can audit: Unfold the model's reasoning trace to see how it reached an answer — or leave it collapsed.
Sources you can open: Inline markers link to a reference list with titles and live links, exactly where the claim is made.
Effort that scales: Choose how hard a model should think, from a quick pass to deep, deliberate reasoning.

Thought FieldOpenAI/GPT-5.5

GPT-5.5

Clarify that it names thoughtlessness, not monstrousness, and tie it to the Eichmann trial. Avoid the common misreading that she was excusing him.

Watching Eichmann on trial, Arendt argued that immense evil can issue not from demonic motive but from sheer thoughtlessness — clichés standing in for judgment[1]. It was never an exoneration: she held him fully responsible, and the phrase names a danger rather than a defense[2], warning how ordinary a person can become once he stops asking what it is he is actually doing.

[1]Eichmann in Jerusalem: A Report on the Banality of Evilarchive.org
[2]Thinking and Moral Considerationsjstor.org

Fig. 2 — A reply, dissected: the thought process unfolded, citations resolved beneath.

IIIYour working memory

It remembers — and reuses.

Two libraries that make the field your own. A memory that quietly learns your context, and a prompt bank that turns your best instructions into one-tap tools — both carried into every conversation, whichever model you open.

A ledger of what matters.

Thought Field distills your preferences, background and habits into prioritized facts — extracted automatically, edited by you, injected into every chat.

Memory Bank

4 memories

Crucial (2)

Writing a dissertation on low-resource machine translation

Prefers answers in English; sources may be bilingual

Important (1)

Works in Python 3.12, with type hints throughout

Normal (1)

Favors concise replies with worked examples

Injected into every conversation — every model already knows your context.

Fig. 3 — The memory ledger: prioritized facts, carried into every conversation.

IVFiles & chats

Bring the whole library.

Two more libraries put your sources within reach. The File Bank holds your uploaded papers and datasets, parsed and ready; the Chat Bank lets you fold a past conversation into a new one. Cue either with a keystroke, mid-sentence.

A shelf of parsed sources.

Upload papers, datasets, slides and scans once. PDFs are read page by page with OCR for scans; Office files are extracted faithfully. Re-attach any of them to a new chat in a keystroke.

/fileCue File

attention-is-all-you-need.pdf

14 pages · OCR applied

Ready

low-resource-mt-survey.pdf

28 pages · text extracted

Ready

results-table.xlsx

Spreadsheet · extracted

Ready

figure-3-attention.png

Image · read by vision

Processing

Fig. 4 — The File Bank: parsed sources, cued mid-sentence with /file.

Every conversation, reusable.

Search your past chats and fold an entire conversation into the current one as context — continuity across threads, without copy-and-paste.

/chatCue Chat

Transformer scaling laws

Claude Opus 4.7

So the compute-optimal ratio suggests…

Multilingual tokenization

GPT-5.5

Byte-level BPE handles rare scripts by…

Evaluation metrics for MT

Gemini 3.1 Pro

BLEU correlates poorly with fluency when…

Positional encoding intuition

DeepSeek V4 Pro

Sinusoidal vs learned embeddings differ when…

Fig. 5 — The Chat Bank: past conversations, folded in with /chat.

VTypesetting

Set like a journal, not a chat log.

Serif headings, measured line lengths, true mathematical notation and properly highlighted code. Long answers read like pages, not bubbles — because you will be reading a great many of them.

latexDisplay math

\operatorname{Attention}(Q,K,V)=\operatorname{softmax}\!\left(\frac{QK^{\top}}{\sqrt{d_k}}\right)V

Fig. 6 — Display mathematics, rendered properly.

python

def attention(q, k, v):
    # scaled dot-product attention
    d_k = q.size(-1)
    scores = q @ k.transpose(-2, -1)
    scores = scores / sqrt(d_k)
    weights = softmax(scores, dim=-1)
    out = weights @ v
    return out

def attention(q, k, v):
    # scaled dot-product attention
    d_k = q.size(-1)
    scores = q @ k.transpose(-2, -1)
    scores = scores / sqrt(d_k)
    weights = softmax(scores, dim=-1)
    out = weights @ v
    return out

Fig. 7 — Code, highlighted and copyable.

Full-text search: Every conversation, searchable in an instant.
Shareable pages: Publish any conversation as a clean, read-only page.
Custom personas: Build assistants with their own instructions and temperament.
Chat presets: Model, reasoning, search and memory, saved as one-tap configurations.
A record of work: Usage statistics and a contribution wall, GitHub-style.
Bilingual throughout: English and 中文, first-class in every corner of the interface.

VIPlans

Start free. Grow when you're ready.

Begin with a generous free tier and upgrade only when your work asks for more. Prices in CNY; pay with Alipay or WeChat, by month, season or year.

Free

Everything you need to start thinking with frontier models.

¥0

Free models only
10 messages / day
No file upload

Start free

Pro Lite

A lighter Pro entry with Basic models always open and a weekly Pro model allowance.

Unlimited access to Basic models
Limited access to Pro models
6 MB file upload
Low reasoning
8 KB memory

Choose Pro Lite

Asked, and answered.

What do I get on the free plan?

Ten messages a day across the free-tier catalogue — DeepSeek, Gemini Flash, Llama and more. No credit card, no trial clock.

Are these really the official models?

Yes. Thought Field calls each laboratory's official API and routes around outages with fallback providers. What you read is what the model actually said.

How does memory work — and can I edit it?

Thought Field distills durable facts from your conversations into a ledger you can review, reprioritize or delete at any time. Memory is injected into new chats automatically.

How do I pay?

Alipay and WeChat Pay, with monthly, seasonal and yearly plans. Redeem codes and invitation rewards are supported as well.

Can I share or export my work?

Any conversation can be published as a read-only link and unpublished just as easily. Your chats, prompts and memory export as JSON.