Summary paragraph on effectiveness and use case; language comparison separated (chart 8, per-model EN/ZH drops in card and report)

README.md

@@ -38,6 +38,28 @@ configs:
 Optical Text In-Context Benchmark: how reliably do LLMs consume text
 delivered as rendered images versus plain text tokens?
 
+In summary, the evaluation reported here finds that optical text compression
+is effective only within a narrow and specific envelope. Delivering content
+as rendered images genuinely reduces input tokens, by sixteen to forty-seven
+per cent across the three frontier models tested, but only when the document
+is long, the rendering is dense and the content is prose that fills the
+page. On short prompts, and on configuration files, tool schemas and source
+code in their natural layout, the image costs between 1.3 and 2.5 times more
+than the text it replaces, so the content that such compression is most
+often deployed on is the content it serves worst. Where the method is
+economical it is also destructive, since exact identifiers, verbatim
+passages and precise figures are misread from dense prose images at rates
+that make them unusable, whilst multi-step reasoning and
+instruction-following largely survive. The damage is consistently larger in
+Chinese than in English for every model tested. No condition measured here
+both saved tokens and preserved exact reading. The appropriate use case is therefore
+bulky, static, prose-heavy context whose gist and structure matter more than
+its exact characters, such as archived conversation history or long
+narrative documents consulted for their sense. The method is unsuitable for
+machine text, for anything that must be recovered verbatim, and for
+latency-sensitive deployments in front of reasoning models, whose thinking
+time on barely legible pages can grow by an order of magnitude.
+
 Text-as-image prompting, as in pxpipe, can pack roughly three
 characters into one vision token versus about one character per text token, a large cost saving whose reliability risks are under-measured. OPTIC-Bench
 measures those risks with paired evaluation: each instance is run under
@@ -102,7 +124,8 @@ The full report is in `RESULTS.md`.
   The models still read roughly ninety per cent of an identifier's
   characters correctly, but a single wrong character fails the field.
   Retrieval from images was also consistently worse in Chinese than in
-  English.
+  English, with a pooled drop of 65 points against 46 on the retrieval
+  families, even though the paired twins share structure and text baselines.
 - Structured machine text is the exception, in both directions. The same
   kinds of value (UUIDs, hashes, tokens) that collapse inside prose survive
   almost intact inside JSON, schemas and code, dropping only four points
@@ -152,6 +175,13 @@ with prose retrieval at the bottom of every panel.
 
 ![Each model's task profile, ranked](charts/7_model_task_cascade.png)
 
+The fifth separates the two languages. Every instance is an English and
+Chinese twin from the same seed, and the text baselines match, so the wider
+Chinese gap is purely an image-channel effect: the same content read from
+pixels loses more in Chinese for every model.
+
+![The image penalty by language](charts/8_language_split.png)
+
 The dense-slice figures come from small samples and are directional. See
 `RESULTS.md` for the per-model core tables, the remaining charts including
 the economy-fidelity trade-off plane, the significance tests, the

RESULTS.md

@@ -208,10 +208,20 @@ characters correctly. How far a structured layout softens this rule is the
 subject of the next subsection.
 
 Retrieval from the image was also consistently worse in Chinese than in
-English. Pooling the three models over the page-scale image conditions, the
-exact-match rate was 0.92 in English against 0.84 in Chinese. This is
-consistent with the greater visual density of Chinese characters making them
-harder to read reliably from a rendered page.
+English, and because every instance is an English and Chinese twin generated
+from the same seed, the comparison is exact rather than incidental. The text
+baselines of the twins match almost perfectly, so the whole difference
+appears in the image channel. Moving content from text to image cost GPT 5.5
+fourteen points in English but twenty-two in Chinese, Gemini 3.5 Flash eight
+against nineteen, and Kimi K2.6 twelve against sixteen. On the two prose
+retrieval families the pooled drop was forty-six points in English and
+sixty-five in Chinese. This is consistent with the greater visual density of
+Chinese characters making them harder to read reliably from a rendered page,
+and it means the economics and the risks of optical compression should be
+assessed per language, since Chinese enjoys the larger token saving whilst
+suffering the larger reading loss.
+
+![The image penalty by language](charts/8_language_split.png)
 
 ### Structured machine text reads back far better than prose