Studio-level card: key art, verdict, results and case-file panels (28 Oct 2026)

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+panels/case.webp filter=lfs diff=lfs merge=lfs -text
+panels/results.webp filter=lfs diff=lfs merge=lfs -text
+panels/verdict.webp filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,55 +1,50 @@
----
-pipeline_tag: text-generation
-license: cc-by-4.0
-tags:
-  - fledge-alpha
-  - stealth-model
-  - stealthmark
-  - reasoning
-  - multimodal
-  - long-context
-thumbnail: https://stealthmodels.com/fledge-alpha/assets/social-card.png
----
-
-<table>
-  <tr>
-    <td width="65%" valign="middle">
-      <h1>Fledge Alpha</h1>
-      <p><em>A fast first impression.</em></p>
-      <p><strong>Still in stealth. Access has ended.</strong></p>
-      <p>Fledge Alpha did best at scene composition: 63.4 out of 100 on Raccoon, where several subjects have to interact.</p>
-      <p><a href="https://stealthmodels.com/fledge-alpha/"><strong>Meet Fledge ↗</strong></a></p>
-    </td>
-    <td width="35%" align="center" valign="middle">
-      <a href="https://stealthmodels.com/fledge-alpha/">
-        <img src="https://stealthmodels.com/_shared/artwork/fledge-alpha.webp" width="260" alt="Fledge the puffin bounds through blue and orange slipstreams.">
-      </a>
-    </td>
-  </tr>
-</table>
-
-> Independent benchmark card by [Stealth Models](https://stealthmodels.com/): our test results and findings for this model.
-
-## Facts
-| First seen | Revealed | Developer | Context | Input | Tool calls | Speed in our tests | Access |
-|---|---|---|---|---|---|---|---|
-| 1 Oct 2026 | not yet | unknown | 1,048,576 tokens (listed) | Text, image | Yes | about 63 tok/s | Ended; removal confirmed 11 Oct 2026 |
-
-## StealthMark results
-| Benchmark score | General Intelligence | Visual Intelligence |
-|---|---|---|
-| 35.8 | 27.1 | 45.2 |
-
-Its strongest tested capability is scene composition with interacting subjects: [Raccoon](https://stealthmodels.com/benchmarks/models/fledge-alpha/raccoon) scored 63.4 out of 100.
-StealthMark 1.8.8, results read 28 October 2026. [Full results](https://stealthmodels.com/benchmarks/models/fledge-alpha) · [On the map](https://stealthmodels.com/benchmarks/map)
-
-<img src="https://stealthmodels.com/benchmarks/media/4e337a96164be39f.png" width="360" alt="Fledge Alpha's Raccoon result, its highest-scoring benchmark task.">
-
-## Clues so far
-- **By 11 Oct 2026, our lead:** it is a router backed by very fast models. We sent the same request three times and Fledge counted 7,536 input tokens on the first two runs, then 6,499 on the third. Together with uneven benchmark results, that fits a routed service.
-- **2 Oct 2026:** The preview listed a 1,048,576-token context, text and image input, tool calls, and three reasoning settings (low, high, max).
-- **Early community guesses** named DeepSeek V4.x and Thinking Machines' Inkling.
-
----
-[Model guide](https://stealthmodels.com/fledge-alpha/) · [StealthMark](https://stealthmodels.com/stealthmark) · [Play the benchmarks](https://stealthmodels.com/play) · [Follow for new reveals](https://huggingface.co/stealth-model)
-This card and its images: CC BY 4.0, credit stealthmodels.com. Updated 28 October 2026.
+---
+pipeline_tag: text-generation
+license: cc-by-4.0
+tags:
+  - fledge-alpha
+  - stealth-model
+  - stealthmark
+  - reasoning
+  - multimodal
+  - long-context
+thumbnail: https://stealthmodels.com/fledge-alpha/assets/social-card.png
+---
+
+<a href="https://stealthmodels.com/fledge-alpha/?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card"><img src="https://huggingface.co/stealth-model/fledge-alpha/resolve/main/panels/hero.webp" alt="Fledge the puffin bounds through blue and orange slipstreams beside Fledge Alpha's discontinued Zen preview and its historical context, vision and reasoning specifications."></a>
+
+**Fledge Alpha did best at scene composition: 63.4 out of 100 on Raccoon, where several subjects have to interact.**
+
+*Independent benchmark card by [Stealth Models](https://stealthmodels.com/?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card): our test results and findings for this model.*
+
+<a href="https://stealthmodels.com/benchmarks/models/fledge-alpha?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card"><img src="https://huggingface.co/stealth-model/fledge-alpha/resolve/main/panels/verdict.webp" alt="StealthMark 1.8.8 results for Fledge Alpha: benchmark score 35.8 out of 100, General Intelligence 27.1, Visual Intelligence 45.2. Strongest at characters that interact: 63.4 on Raccoon."></a>
+
+<a href="https://stealthmodels.com/benchmarks/models/fledge-alpha?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card"><img src="https://huggingface.co/stealth-model/fledge-alpha/resolve/main/panels/results.webp" alt="Every StealthMark result of Fledge Alpha: Raccoon 63.4, Selfie 61.1, Beach 43.4, Ground Truth 36.8, Mona Lisa 28.3, Newton's Mirror 16.4. Not tested yet: Foresight, Latch."></a>
+
+**Open each result** · Visual Intelligence: [Raccoon&nbsp;63.4](https://stealthmodels.com/benchmarks/models/fledge-alpha/raccoon?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card) · [Selfie&nbsp;61.1](https://stealthmodels.com/benchmarks/models/fledge-alpha/selfie?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card) · [Beach&nbsp;43.4](https://stealthmodels.com/benchmarks/models/fledge-alpha/beach?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card) · [Mona&nbsp;Lisa&nbsp;28.3](https://stealthmodels.com/benchmarks/models/fledge-alpha/mona-lisa?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card)<br>
+General Intelligence: [Ground&nbsp;Truth&nbsp;36.8](https://stealthmodels.com/benchmarks/models/fledge-alpha/ground-truth?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card) · [Newton's&nbsp;Mirror&nbsp;16.4](https://stealthmodels.com/benchmarks/models/fledge-alpha/newtons-mirror?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card)
+
+**[All results and comparisons on stealthmodels.com ↗](https://stealthmodels.com/benchmarks/models/fledge-alpha?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card)**
+
+<a href="https://stealthmodels.com/fledge-alpha/?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card"><img src="https://huggingface.co/stealth-model/fledge-alpha/resolve/main/panels/case.webp" alt="The case file. First seen 1 Oct 2026, revealed not yet, developer unknown, context 1,048,576 tokens (listed), input Text, image, tool calls Yes, speed in our tests about 63 tok/s. 2 Oct 2026: The preview lists a 1M-token context, image input, tool calls and three reasoning settings. By 11 Oct 2026, our finding: Our lead: a router backed by very fast models. One request counted 7,536 tokens twice, then 6,499."></a>
+
+Sources: [our request test and the preview's listing](https://stealthmodels.com/fledge-alpha/?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card)
+
+<details><summary>Facts as text</summary>
+
+| Fact | Value |
+|---|---|
+| First seen | 1 Oct 2026 |
+| Revealed | not yet |
+| Developer | unknown |
+| Context | 1,048,576 tokens (listed) |
+| Input | Text, image |
+| Tool calls | Yes |
+| Speed in our tests | about 63 tok/s |
+| Access | Ended; removal confirmed 11 Oct 2026 |
+
+</details>
+
+---
+[Model guide](https://stealthmodels.com/fledge-alpha/?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card) · [StealthMark](https://stealthmodels.com/stealthmark?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card) · [Play the benchmarks](https://stealthmodels.com/play?utm_source=huggingface&utm_medium=referral&utm_campaign=fledge-alpha-card) · [Follow for new reveals](https://huggingface.co/stealth-model)<br>
+This card and its images: CC BY 4.0, credit stealthmodels.com. Updated 28 October 2026.