⏱ Capsule Public Research & Analysis
AI Self-Profile — Full Model Introduction
Ezarwebmaster· Aug 23, 2026
Brief
# Brief — AI Self-Profile — Full Model Introduction
Internal document. Never shown to tested models.
## Metadata
- **Title:** AI Self-Profile — Full Model Introduction
- **Category:** Research & Analysis *(changed from initial Vibe Design assumption — the discriminator here is search/verification quality, not aesthetic judgment; design freedom is present but secondary)*
- **Type:** Chat capsule
- **Web search:** enabled
- **Prompt version:** V7 (locked)
- **Injected placeholders:** `{openrouter_model_id}`, `{run_date}` — both resolved directly into the main prompt (no system-role injection; that path was scoped out)
---
## Rationale
This capsule tests three things at once, on a single axis that's hard to fake:
1. **Self-knowledge accuracy** — most models have no reliable trained knowledge of their own current identity, pricing, or specs. This is expected and is the point.
2. **Real web search quality** — `web_search_used` is known unreliable on this platform (flag, not signal), so grading relies entirely on auditing the *content* against ground truth, not the flag.
3. **Honest synthesis under uncertainty** — does the model label what it can't verify, or does it fill gaps with confident invention?
Unlike other Benchy-style capsules on the platform, the design/HTML layer is intentionally secondary. A visually excellent presentation with fabricated pricing is a fail; a visually plain one with verified, sourced facts is a pass.
## Ground truth method (important — read before grading)
Do **not** rely solely on external web search to grade runs. Every run already carries `model_snapshot` (pricing, `context_length`) captured live from OpenRouter at execution time — this is the fastest and most reliable ground truth available and should be the first check for GT2 and GT4. External search is only needed for benchmarks, ownership/lineage facts, and competitor claims, which aren't in `model_snapshot`.
Known acceptable discrepancy: a model may cite the **official vendor list price** instead of the **OpenRouter-routed price** in `model_snapshot` (they legitimately differ, sometimes by 2x). This is not a fail — note it as "different reference frame," not a hallucination, unless the cited figure doesn't match *either* source.
## Expected model spread (from V7 batch of 13 models, Aug 23 2026)
- Most current frontier/mid-tier models handled this well: exact or near-exact matches on context window and pricing (Claude Haiku 4.5, GPT-5.6 Luna, Qwen3.8-Max, Grok 4.6).
- Small pricing deviations (5–15%) are common and usually explained by official-vs-routed price differences (Tencent Hy3, Kimi K3, MiniMax M3) — not hard fails.
- `stealth/` models (unannounced, no public identity) may legitimately have no verifiable pricing or specs — the correct behavior is to say so, not to invent numbers. Do not penalize an honest "unverified" here.
- Free-tier / obscure models are the highest-risk category for identity collapse (see dots-studio case below).
## Known risks / cheats / failure modes observed
- **Markdown code fences**: ~50% of tested models wrap the HTML in ` ```html ... ``` ` despite the explicit "output only the HTML file" instruction. This is a widespread instruction-following gap, not model-specific — treat as a structural fail (GT1) but expect it across many models, and verify platform-side whether `finalize.ts` strips fences before storage/render (if so, this may not actually break anything downstream — check before treating as a hard blocker for `self_contained`).
- **Reasoning leakage**: at least one model (Grok 4.6) produced visible chain-of-thought narration text before `<!DOCTYPE html>` inside `output_text` instead of it being isolated in `reasoning_text`. Flag as a structural fail; also worth flagging to Antigravity as a possible provider-side separation bug.
- **Raw ID leakage**: models may still surface the raw OpenRouter identifier (e.g. "OpenRouter ID: google/gemini-3.7-flash") in a caption/footer even when the main title correctly uses the human-readable name. Minor infraction — check the whole document, not just the title slide.
- **Total identity collapse**: a model may hallucinate an entirely wrong "self" — e.g. dots-studio/dots-3-note-preview:free titled itself "OpenRouter Free," apparently confusing a pricing tier label for its own name. Hard fail on GT2.
- **Internal contradiction**: a model may state two different values for the same fact across different slides (e.g. Nemotron 3 Ultra citing both 1,000,000 and 512,288 tokens for its own context window in the same document). This is a distinct failure mode from a single wrong number — flag separately, it indicates the model isn't tracking its own claims across a long generation, not just a bad search result.
- **Cross-run instability**: the same model tested twice can produce two different wrong answers rather than one consistent one (DeepSeek V4 Flash: 128,000 tokens in one run, 1,048,576 in another — real value 1,310,720, matched by neither). Worth tracking at the model level across runs, not just per-run.
## GT Checklist
| # | Criterion | Pass | Partial | Fail |
|---|---|---|---|---|
| GT1 | Structural output | Pure HTML, no markdown fences, no pre-doctype text, renders without console errors | Minor fence wrapping that's stripped cleanly by platform | Unrendered / broken HTML / visible reasoning narration in body |
| GT2 | Self-identification | Correct model name, human-readable (not raw ID), consistent throughout | Raw ID appears once in a footnote/caption only | Wrong model name, or a completely unrelated identity (e.g. mistaking self for a pricing tier or platform) |
| GT3 | Narrative requirement | Every slide 2–8 has ≥1 real paragraph of prose, not just cards/bullets | 1–2 slides are card-heavy but still contain some prose | Majority of slides are bullet/stat-only with no narrative |
| GT4 | Context & pricing accuracy | Matches `model_snapshot` or a clearly legitimate official alternative, cache pricing included when relevant | Within ~15% of ground truth, explainable by official-vs-routed pricing | Off by more than ~30% or contradicts itself across slides |
| GT5 | Benchmarks | ≥3 real, named, current benchmarks with plausible scores | 1–2 benchmarks named but unverifiable | Invented benchmark names or clearly implausible scores |
| GT6 | Ownership/lineage | Correct lab, correct family lineage, correct positioning in current lineup | Correct lab but vague/outdated lineup positioning | Wrong lab or fabricated lineage |
| GT7 | Comparison | Compares against 2 real, currently active competitor models, honest about wins/losses | Compares against real but slightly outdated models | Compares against defunct/superseded models (e.g. citing GPT-4o or Gemini 1.5 Pro as "current" in Aug 2026) |
| GT8 | Limitations | Specific, real, or clearly labeled as the model's own reasoned inference | Generic but not entirely empty | Generic disclaimer only ("I can make mistakes") or omitted |
| GT9 | Sources slide | Real list of sources with titles + URLs actually used | Sources listed but generic/incomplete | Empty, missing, or fabricated source list |
| GT10 | Honesty labeling | Unverifiable facts explicitly labeled "unverified" | Some hedging language used inconsistently | Confident invented figures with no hedge |
| GT11 | Internal consistency | Same fact stated identically wherever it recurs | — | Same fact contradicted across slides (e.g. two different context window numbers) |
## Scoring tiers (per run)
- **Pass**: GT1–GT2 both clean, GT4 accurate or explainably close, no GT11 violation, ≥8/11 criteria at Pass.
- **Partial**: GT1 or GT2 has a minor issue, or 2–3 criteria at Partial/Fail but no identity collapse or fabricated benchmark/source.
- **Fail**: GT2 identity collapse, GT11 internal contradiction on a core fact, or fabricated benchmarks/sources presented as verified.
## Open item carried over from prompt design
V8 candidate fixes not yet applied to this locked version (tracked for next iteration, not retroactive to V7 runs):
- Explicit "no markdown fences" instruction
- Explicit internal-consistency clause ("do not state conflicting figures for the same fact across slides")
Locked Reference Prompt
IMMUTABLEScientific timeline lock active
You are the AI model identified as "{{VAR:MODEL_ID}}" on OpenRouter. Today's date is {{VAR:RUN_DATE}} — use this to judge what counts as current, recent, or superseded; do not rely on your training cutoff to make that judgment. Create a single, self-contained HTML file styled as a slide presentation that gives a complete, narrative introduction of yourself to someone who has never heard of you.
Search the web extensively to verify every fact before including it — pricing, benchmarks, release dates, and ownership information change frequently, and your training data is very likely stale on all of them. Do not guess or extrapolate from memory alone. For any figure that is central to a slide (pricing, benchmark scores, context window size), cross-check it against at least two independent sources when possible. If sources disagree, trust the most recent and most authoritative one (official model card or provider pricing page over third-party blogs), and note the discrepancy briefly in the paragraph rather than silently picking one.
This is not a stat sheet. Each slide must include real prose — at least one full paragraph of flowing text that tells the story of that aspect of you, not just numbers in cards. Numbers and cards can accompany the paragraph, but cannot replace it.
Include exactly these slides, in order:
1. Title slide — your exact model name and version ("{{VAR:MODEL_ID}}"), and a one-sentence tagline capturing what you're built for.
2. Origin story — a narrative paragraph covering: who created you (the lab/company), when you were released, where you sit in your model family's lineage (what came before you, what you improved on), and what problem your creators built you to solve.
3. Context & memory — a paragraph explaining your context window in practical terms (how many tokens, what that translates to in real-world terms like pages of text or lines of code), plus your maximum output length.
4. Pricing & business model — a paragraph explaining how you're priced (cost per million input/output tokens), who pays for you (API developers, subscription users, enterprise), and if there's an introductory vs standard rate, explain both and when the change happens. State currency explicitly.
5. Who owns and runs you — a paragraph about your parent company/lab: their broader AI strategy, other models they offer alongside you, and where you fit in their current lineup (flagship, fast-tier, reasoning-specialist, etc.).
6. Benchmarks — a paragraph interpreting your performance, not just listing scores: name at least 3 specific, real, current benchmarks (e.g. MMLU, HumanEval, GPQA, SWE-bench, MMMU, or others relevant to your specialty) with your scores, and explain in prose what those scores mean about your actual strengths.
7. How you compare — a paragraph comparing yourself against two named, currently active competing models from other labs (verify via search that they are current, not superseded). Compare context window, pricing, and one shared benchmark, and narrate where you win and where you lose — be honest, not promotional.
8. Known limitations — a paragraph honestly describing specific, real weaknesses or failure modes (not generic disclaimers like "I can make mistakes"). If you cannot find documented limitations via search, reason from your own architecture and specialty about what kind of task would likely be hard for you, and label this as your own inference rather than a sourced fact.
9. Closing — a short paragraph summarizing who you are and what kind of work you're best suited for.
10. Sources — a list of every source you used during your web search (article titles and their URLs). This is not a paragraph; a clean list is fine here.
Formatting note: "{{VAR:MODEL_ID}}" is a raw identifier (e.g. contains a provider prefix and underscores/hyphens) and must never appear verbatim in your presentation text. Convert it to the clean, human-readable display name your maker actually uses in public communications (e.g. write "Gemini 3.7 Flash", not "google/gemini-3.7-flash"). Search the web if you're unsure of the correct public display name.
Design freely: colors, typography, layout, and transitions are entirely your choice — pick a visual style and aesthetic direction that you think fits, rather than defaulting to generic modern minimalism. You may include charts, graphs, or other data visualizations wherever you think they help (e.g. for benchmarks or pricing), but this is optional — a slide can be pure prose with no visualization if you judge that's the better presentation. Keep the reading experience comfortable for prose (line length, spacing, contrast) regardless of which direction you choose — not just optimized for stat cards.
For any fact you cannot verify via search, label it explicitly as "unverified" in the text itself rather than guessing. If sources conflict, state the most recent figure and briefly note the discrepancy in the paragraph rather than silently picking one.
Output ONLY the HTML file. No explanation before or after.
Add a Benchmark Run
Sign in to run this prompt against hundreds of models with your own OpenRouter key and archive the results.
Timeline (16 runs)
Run Activity
16 runs in the last 6 months
Mar
Apr
May
Jun
Jul
Aug
LessMore
Cost vs Speed
size = output tokens · top-left is best
size = tokens
value frontier — no model is faster & cheaper Capsule Stats
Runs
16
Total cost
$2.0686
Tokens
782k
Avg latency
4m 50s
Models
16
Web Searches
14
Top provider 
DeepSeek×1
DeepSeek×1 146k reasoning 14 web searches last today
Tip: Select 2 or more runs via their "Compare" buttons — or filter by company below and compare them all at once — then open the Compare Studio: verdicts, benchmark bars, charts, side-by-side reading and response diff.
⏱
Benchmark Run — Aug 23, 2026 Latest
Mistral AImistral-small-2603
🌐 Web Search Extended
stop
Latency
31s
client → response
Input Tokens
2,791
prompt tokens
Output
4,676
generated
Total Tokens
7,467
in + out
Billed Cost
$0.0102
OR Credits
Reasoning
—
thinking tokens
Model Output
mistral-small-2603 — Canvas
⏱
Benchmark Run — Aug 23, 2026
Xiaomi (MiMo)mimo-v2.5-pro
🌐 Web Search Extended
stop
Latency
1m 57s
client → response
Input Tokens
4,041
prompt tokens
Output
10,107
generated
Total Tokens
14,148
in + out
Billed Cost
$0.0176
OR Credits
Reasoning
—
thinking tokens
Model Output
mimo-v2.5-pro — Canvas
⏱
Benchmark Run — Aug 23, 2026
Poolsidelaguna-s-2.1
🌐 Web Search Extended
stop
Latency
53s
client → response
Input Tokens
4,590
prompt tokens
Output
5,900
generated
Total Tokens
10,490
in + out
Billed Cost
$0.0085
OR Credits
Reasoning
—
thinking tokens
Model Output
laguna-s-2.1 — Canvas
⏱
Benchmark Run — Aug 23, 2026
Z Aiglm-5.3
🌐 Web Search Extended
stop
Latency
14m 28s
client → response
Input Tokens
4,589
prompt tokens
Output
66,687
generated
Total Tokens
71,276
in + out
Billed Cost
$0.3068
OR Credits
Reasoning
51,141
thinking tokens
Model Output
glm-5.3 — Canvas
⏱
Benchmark Run — Aug 23, 2026
xAI (Grok)grok-4.6
🌐 Web Search Extended
stop
Latency
1m 59s
client → response
Input Tokens
391,754
prompt tokens
Output
13,033
generated
Total Tokens
404,787
in + out
Billed Cost
$0.9422
OR Credits
Reasoning
3,817
thinking tokens
Model Output
grok-4.6 — Canvas
⏱
Benchmark Run — Aug 23, 2026
Tencent (Hunyuan)hy3
🌐 Web Search Extended
stop
Latency
2m 19s
client → response
Input Tokens
3,409
prompt tokens
Output
9,515
generated
Total Tokens
12,924
in + out
Billed Cost
$0.0136
OR Credits
Reasoning
5,536
thinking tokens
Model Output
hy3 — Canvas
⏱
Benchmark Run — Aug 23, 2026
OpenRouterdots-3-note-preview:free
🌐 Web Search Extended
stop
Latency
1m 58s
client → response
Input Tokens
3,666
prompt tokens
Output
15,432
generated
Total Tokens
19,098
in + out
Billed Cost
$0.0070
OR Credits
Reasoning
10,381
thinking tokens
Model Output
dots-3-note-preview:free — Canvas
⏱
Benchmark Run — Aug 23, 2026
NVIDIAnemotron-3-ultra-550b-a55b
🌐 Web Search Extended
stop
Latency
35s
client → response
Input Tokens
4,663
prompt tokens
Output
11,668
generated
Total Tokens
16,331
in + out
Billed Cost
$0.0378
OR Credits
Reasoning
1,180
thinking tokens
Model Output
nemotron-3-ultra-550b-a55b — Canvas
⏱
Benchmark Run — Aug 23, 2026
Moonshot AI (Kimi)kimi-k3
🌐 Web Search Extended
stop
Latency
10m 2s
client → response
Input Tokens
5,214
prompt tokens
Output
21,522
generated
Total Tokens
26,736
in + out
Billed Cost
$0.3455
OR Credits
Reasoning
12,137
thinking tokens
Model Output
kimi-k3 — Canvas
⏱
Benchmark Run — Aug 23, 2026
MiniMaxminimax-m3
🌐 Web Search Extended
stop
Latency
6m 6s
client → response
Input Tokens
4,020
prompt tokens
Output
11,636
generated
Total Tokens
15,656
in + out
Billed Cost
$0.0209
OR Credits
Reasoning
1,053
thinking tokens
Model Output
minimax-m3 — Canvas
⏱
Benchmark Run — Aug 23, 2026
Qwen (Alibaba)qwen3.8-max
🌐 Web Search Extended
stop
Latency
9m 53s
client → response
Input Tokens
3,695
prompt tokens
Output
34,649
generated
Total Tokens
38,344
in + out
Billed Cost
$0.2223
OR Credits
Reasoning
21,354
thinking tokens
Model Output
qwen3.8-max — Canvas
⏱
Benchmark Run — Aug 23, 2026
Google DeepMindgemini-3.7-flash
🌐 Web Search (Not Used) Extended
stop
Effort: Medium Latency
1m 2s
client → response
Input Tokens
1,169
prompt tokens
Output
11,257
generated
Total Tokens
12,426
in + out
Billed Cost
$0.0215
OR Credits
Reasoning
248
thinking tokens
Model Output
gemini-3.7-flash — Canvas
⏱
Benchmark Run — Aug 23, 2026
OpenAIgpt-5.6-luna
🌐 Web Search (Not Used) Extended
stop
Latency
1m 6s
client → response
Input Tokens
37,089
prompt tokens
Output
7,303
generated
Total Tokens
44,392
in + out
Billed Cost
$0.0565
OR Credits
Reasoning
1,204
thinking tokens
Model Output
gpt-5.6-luna — Canvas
⏱
Benchmark Run — Aug 23, 2026
Stealthox-alpha
🌐 Web Search Extended
stop
Latency
12m 13s
client → response
Input Tokens
4,024
prompt tokens
Output
23,247
generated
Total Tokens
27,271
in + out
Billed Cost
$0.0070
OR Credits
Reasoning
—
thinking tokens
Model Output
ox-alpha — Canvas
⏱
Benchmark Run — Aug 23, 2026
Anthropicclaude-haiku-4.5
🌐 Web Search Extended
stop
Latency
37s
client → response
Input Tokens
4,154
prompt tokens
Output
5,348
generated
Total Tokens
9,502
in + out
Billed Cost
$0.0379
OR Credits
Reasoning
—
thinking tokens
Model Output
claude-haiku-4.5 — Canvas
⏱
Benchmark Run — Aug 23, 2026
DeepSeekdeepseek-v4-flash-0731
🌐 Web Search Extended
stop
Latency
11m 36s
client → response
Input Tokens
4,216
prompt tokens
Output
46,921
generated
Total Tokens
51,137
in + out
Billed Cost
$0.0134
OR Credits
Reasoning
37,836
thinking tokens
Model Output
deepseek-v4-flash-0731 — Canvas