## Chart Type: Comparative Line Graphs of Translation Performance ### Overview The image presents six line graphs comparing the performance of three language models (Gemini 1.5 Pro, Gemini 1.5 Flash, and GPT-4 Turbo) on English translation tasks into different target languages. The x-axis represents the number of shots (K), which is the number of examples used for training, and the y-axis represents the Test chrF score, a metric for translation quality. The graphs are arranged in a 2x3 grid, each focusing on a different target language. ### Components/Axes * **Title:** Comparative Line Graphs of Translation Performance * **Legend:** Located at the top of the image. * Gemini 1.5 Pro (Yellow-Green) * Gemini 1.5 Flash (Blue) * GPT-4 Turbo (Light Gray) * **X-axis:** Number of Shots (K). Logarithmic scale with base 2. Markers at 2^0, 2^1, 2^2, 2^3, 2^4, 2^5, 2^6, 2^7, 2^8, 2^9, 2^10, 2^11, 2^12. * **Y-axis:** Test chrF. * Top row (Flores): Scale from 30.0 to 50.0, with ticks at 30.0, 35.0, 40.0, 45.0, 50.0. * Bottom row (In-house): Scale from 10.0 to 35.0, with ticks at 10.0, 15.0, 20.0, 25.0, 30.0, 35.0. * **Titles of Subplots:** * Translation: English -> Bemba * Translation: English -> Kurdish * Translation: English -> Ewe * Translation: English -> Acholi * Translation: English -> Abkhaz * Translation: English -> Navajo ### Detailed Analysis **Graph 1: English -> Bemba** * Y-axis: Test chrF (Flores) * Gemini 1.5 Pro (Yellow-Green): Starts at approximately 40.5 and increases to approximately 50.5. * Gemini 1.5 Flash (Blue): Starts at approximately 23.5 and increases to approximately 41.0. * GPT-4 Turbo (Light Gray): Starts at approximately 30.0 and increases to approximately 38.0. **Graph 2: English -> Kurdish** * Y-axis: Test chrF (Flores) * Gemini 1.5 Pro (Yellow-Green): Starts at approximately 45.0 and increases slightly to approximately 46.5. * Gemini 1.5 Flash (Blue): Starts at approximately 42.5 and increases slightly to approximately 44.5. * GPT-4 Turbo (Light Gray): Starts at approximately 35.0 and fluctuates slightly, ending around 32.0. **Graph 3: English -> Ewe** * Y-axis: Test chrF (Flores) * Gemini 1.5 Pro (Yellow-Green): Starts at approximately 41.5 and increases slightly to approximately 43.0. * Gemini 1.5 Flash (Blue): Starts at approximately 30.0 and increases to approximately 37.5. * GPT-4 Turbo (Light Gray): Starts at approximately 24.0 and increases to approximately 27.0. **Graph 4: English -> Acholi** * Y-axis: Test chrF (In-house) * Gemini 1.5 Pro (Yellow-Green): Starts at approximately 33.5 and increases slightly to approximately 35.5. * Gemini 1.5 Flash (Blue): Starts at approximately 17.0, dips to approximately 13.0, and then increases to approximately 30.0. * GPT-4 Turbo (Light Gray): Starts at approximately 22.0 and fluctuates slightly, ending around 24.0. **Graph 5: English -> Abkhaz** * Y-axis: Test chrF (In-house) * Gemini 1.5 Pro (Yellow-Green): Starts at approximately 28.0 and increases slightly to approximately 29.0. * Gemini 1.5 Flash (Blue): Starts at approximately 5.0 and increases to approximately 32.0. * GPT-4 Turbo (Light Gray): Starts at approximately 30.0 and fluctuates slightly, ending around 28.0. **Graph 6: English -> Navajo** * Y-axis: Test chrF (In-house) * Gemini 1.5 Pro (Yellow-Green): Starts at approximately 25.0 and increases to approximately 34.0. * Gemini 1.5 Flash (Blue): Starts at approximately 11.0, dips to approximately 9.0, and then increases to approximately 28.0. * GPT-4 Turbo (Light Gray): Starts at approximately 20.0 and fluctuates slightly, ending around 22.0. ### Key Observations * Gemini 1.5 Pro consistently performs well across all translation tasks, generally achieving the highest chrF scores. * Gemini 1.5 Flash shows significant improvement with an increasing number of shots, particularly in the "In-house" datasets (Acholi, Abkhaz, Navajo). * GPT-4 Turbo's performance is relatively stable across different numbers of shots, but it generally scores lower than Gemini 1.5 Pro and, in many cases, lower than Gemini 1.5 Flash with sufficient shots. * The "Flores" datasets (Bemba, Kurdish, Ewe) generally yield higher chrF scores compared to the "In-house" datasets (Acholi, Abkhaz, Navajo). * The performance of Gemini 1.5 Flash on Acholi and Navajo translation tasks initially decreases before increasing with more shots. ### Interpretation The data suggests that Gemini 1.5 Pro is a strong performer for English translation across various languages, demonstrating high accuracy even with limited training examples. Gemini 1.5 Flash benefits significantly from increased training data, eventually approaching or even surpassing GPT-4 Turbo's performance in some cases. GPT-4 Turbo provides a baseline level of performance that is relatively consistent regardless of the number of training examples. The difference in chrF scores between the "Flores" and "In-house" datasets may indicate variations in the complexity or quality of the datasets themselves. The initial dip in performance for Gemini 1.5 Flash on Acholi and Navajo translation tasks could be due to overfitting or other optimization challenges at lower shot counts. Overall, the graphs highlight the importance of model selection and training data size for achieving optimal translation performance. Gemini 1.5 Pro appears to be a robust choice, while Gemini 1.5 Flash can be a viable alternative with sufficient training data.

## Chart: Translation Performance Comparison ### Overview This image presents a series of six line charts comparing the performance of three language models – Gemini 1.5 Pro, Gemini 1.5 Flash, and GPT-4 Turbo – on English-to-various-language translation tasks. The performance metric is Test chrF (a measure of translation quality), plotted against the Number of Shots (K), representing the number of example translations provided to the model. Each chart focuses on a specific target language. ### Components/Axes * **X-axis:** Number of Shots (K), ranging from 2⁰ to 2¹² (approximately 1 to 4096). The axis is labeled "Number of Shots (K)". * **Y-axis:** Test chrF score. The top three charts have a Y-axis labeled "Test chrF (Flores)" with a scale from approximately 25 to 50. The bottom three charts have a Y-axis labeled "Test chrF (In-house)" with a scale from approximately 15 to 35. * **Lines:** Three lines representing the performance of: * Gemini 1.5 Pro (Yellow) * Gemini 1.5 Flash (Blue) * GPT-4 Turbo (Light Blue) * **Legend:** Located at the top-center of the image, clearly labeling each line with its corresponding model name and color. * **Titles:** Each sub-chart has a title indicating the translation direction: "Translation: English → [Language]". The languages are: Bemba, Kurdish, Ewe, Acholi, Abkhaz, and Navajo. ### Detailed Analysis or Content Details **Chart 1: English → Bemba** * Gemini 1.5 Pro: The line is relatively flat, starting at approximately 41.5 and ending at approximately 42.5. * Gemini 1.5 Flash: The line slopes upward, starting at approximately 35 and ending at approximately 45. * GPT-4 Turbo: The line slopes upward, starting at approximately 38 and ending at approximately 43. **Chart 2: English → Kurdish** * Gemini 1.5 Pro: The line is relatively flat, starting at approximately 43 and ending at approximately 44. * Gemini 1.5 Flash: The line slopes upward, starting at approximately 35 and ending at approximately 45. * GPT-4 Turbo: The line slopes upward, starting at approximately 38 and ending at approximately 42. **Chart 3: English → Ewe** * Gemini 1.5 Pro: The line is relatively flat, starting at approximately 32 and ending at approximately 33. * Gemini 1.5 Flash: The line slopes upward, starting at approximately 25 and ending at approximately 38. * GPT-4 Turbo: The line slopes upward, starting at approximately 28 and ending at approximately 35. **Chart 4: English → Acholi** * Gemini 1.5 Pro: The line is relatively flat, starting at approximately 28 and ending at approximately 29. * Gemini 1.5 Flash: The line slopes upward, starting at approximately 20 and ending at approximately 30. * GPT-4 Turbo: The line slopes downward, starting at approximately 25 and ending at approximately 18. **Chart 5: English → Abkhaz** * Gemini 1.5 Pro: The line slopes downward, starting at approximately 32 and ending at approximately 25. * Gemini 1.5 Flash: The line slopes upward, starting at approximately 20 and ending at approximately 30. * GPT-4 Turbo: The line is relatively flat, starting at approximately 25 and ending at approximately 26. **Chart 6: English → Navajo** * Gemini 1.5 Pro: The line is relatively flat, starting at approximately 24 and ending at approximately 25. * Gemini 1.5 Flash: The line slopes downward, starting at approximately 28 and ending at approximately 20. * GPT-4 Turbo: The line slopes upward, starting at approximately 18 and ending at approximately 28. ### Key Observations * Gemini 1.5 Pro generally exhibits a stable performance across all languages, with minimal improvement as the number of shots increases. * Gemini 1.5 Flash consistently shows improvement in chrF score as the number of shots increases, across all languages. * GPT-4 Turbo's performance varies significantly depending on the target language. It shows improvement for some languages (Kurdish, Ewe, Navajo) but declines for others (Acholi). * The "In-house" chrF scores (Charts 4-6) are generally lower than the "Flores" chrF scores (Charts 1-3). * The largest performance gains from increasing the number of shots are observed for Gemini 1.5 Flash. ### Interpretation The data suggests that Gemini 1.5 Pro benefits less from increasing the number of example translations (shots) compared to Gemini 1.5 Flash and GPT-4 Turbo. This could indicate that Gemini 1.5 Pro has a stronger inherent understanding of translation principles or a more robust internal representation of language. Gemini 1.5 Flash consistently improves with more shots, suggesting it is more data-driven and benefits from learning from examples. GPT-4 Turbo's performance is more language-specific, potentially indicating variations in the quality of training data or the complexity of the translation task for each language. The difference between "Flores" and "In-house" chrF scores suggests that the evaluation datasets have different characteristics, potentially reflecting different translation styles or domains. The charts provide a comparative analysis of translation quality across different models and languages, highlighting the strengths and weaknesses of each approach. The varying trends suggest that the optimal model choice may depend on the specific language pair and the availability of example translations.

\n ## Multi-Panel Line Chart: AI Model Translation Performance (chrF Score vs. Number of Shots) ### Overview The image displays a set of six line charts arranged in a 2x3 grid. Each chart compares the translation performance of three large language models—Gemini 1.5 Pro, Gemini 1.5 Flash, and GPT-4 Turbo—from English to a specific target language. Performance is measured using the chrF score (a character-level F-score for machine translation evaluation) as a function of the number of "shots" (in-context learning examples), plotted on a logarithmic scale (base 2). The overall trend shows that performance generally improves with more shots for all models, but the rate of improvement and absolute performance vary significantly by language and model. ### Components/Axes * **Legend:** Positioned at the top center of the entire figure. It defines the three data series: * **Gemini 1.5 Pro:** Represented by a solid, light green line with circular markers. * **Gemini 1.5 Flash:** Represented by a solid, light blue line with circular markers. * **GPT-4 Turbo:** Represented by a solid, light gray line with circular markers. * **X-Axis (Common to all subplots):** Labeled "Number of Shots (K)". The scale is logarithmic base 2, with tick marks at 2⁰, 2¹, 2², 2³, 2⁴, 2⁵, 2⁶, 2⁷, 2⁸, 2⁹, 2¹⁰, 2¹¹, and 2¹² (representing 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, and 4096 shots, respectively). * **Y-Axis (Varies by subplot):** Labeled "Test chrF (Flores)" for the top row and "Test chrF (In-house)" for the bottom row. The numerical scale and range differ for each language chart. * **Subplot Titles:** Each of the six charts has a title indicating the translation task: "Translation: English → [Target Language]". The languages are, from top-left to bottom-right: Bemba, Kurdish, Ewe, Acholi, Abkhaz, and Navajo. ### Detailed Analysis (Per Subplot) **1. Top-Left: English → Bemba** * **Y-Axis Range:** ~25.0 to 50.0. * **Trends & Key Points:** * **Gemini 1.5 Pro (Green):** Starts highest at ~40.5 (2⁰ shots). Shows a steady, strong upward trend, reaching the highest overall score on the chart of ~50.0 at 2¹² shots. * **Gemini 1.5 Flash (Blue):** Starts lowest at ~24.0 (2⁰ shots). Exhibits the steepest improvement curve, surpassing GPT-4 Turbo around 2⁶ shots and ending at ~44.0 at 2¹² shots. * **GPT-4 Turbo (Gray):** Starts at ~29.5 (2⁰ shots). Shows a moderate, steady increase, ending at ~37.0 at 2¹² shots. It is overtaken by Gemini 1.5 Flash mid-chart. **2. Top-Center: English → Kurdish** * **Y-Axis Range:** ~35.0 to 47.0. * **Trends & Key Points:** * **Gemini 1.5 Pro (Green):** Consistently the top performer. Starts at ~45.0 (2⁰ shots) and shows a gentle upward slope, ending at ~46.5 at 2¹² shots. * **Gemini 1.5 Flash (Blue):** Starts at ~42.5 (2⁰ shots). Follows a similar gentle upward trend parallel to Gemini 1.5 Pro, ending at ~45.0 at 2¹² shots. * **GPT-4 Turbo (Gray):** Starts at ~35.5 (2⁰ shots). Performance is relatively flat with minor fluctuations, ending at ~35.5 at 2¹² shots. It shows minimal improvement with more shots. **3. Top-Right: English → Ewe** * **Y-Axis Range:** ~25.0 to 43.0. * **Trends & Key Points:** * **Gemini 1.5 Pro (Green):** Dominant performer. Starts at ~42.0 (2⁰ shots) and remains nearly flat, with a very slight upward trend to ~43.0 at 2¹² shots. * **Gemini 1.5 Flash (Blue):** Starts at ~29.5 (2⁰ shots). Shows a strong, consistent upward trend, ending at ~37.5 at 2¹² shots. * **GPT-4 Turbo (Gray):** Starts lowest at ~24.0 (2⁰ shots). Shows a steady but slower upward trend compared to Gemini 1.5 Flash, ending at ~27.5 at 2¹² shots. **4. Bottom-Left: English → Acholi** * **Y-Axis Range:** ~15.0 to 35.0. * **Trends & Key Points:** * **Gemini 1.5 Pro (Green):** Top performer. Starts at ~33.0 (2⁰ shots) and shows a very gradual increase, ending at ~35.0 at 2¹² shots. * **Gemini 1.5 Flash (Blue):** Exhibits a notable dip. Starts at ~17.0 (2⁰ shots), drops to a low of ~13.0 at 2⁵ shots, then recovers sharply, ending at ~29.0 at 2¹² shots. * **GPT-4 Turbo (Gray):** Starts at ~21.5 (2⁰ shots). Shows a very slow, steady increase, ending at ~23.0 at 2¹² shots. **5. Bottom-Center: English → Abkhaz** * **Y-Axis Range:** ~5.0 to 35.0. * **Trends & Key Points:** * **Gemini 1.5 Pro (Green):** Starts at ~25.5 (2⁰ shots). Performance is flat until 2⁸ shots, then increases sharply, ending at ~34.0 at 2¹² shots. * **Gemini 1.5 Flash (Blue):** Starts very low at ~7.0 (2⁰ shots). Remains flat until 2⁶ shots, then begins a dramatic, steep climb, ending at ~28.0 at 2¹² shots. * **GPT-4 Turbo (Gray):** Starts at ~29.0 (2⁰ shots), making it the initial leader. Performance is flat with a slight dip, ending at ~28.0 at 2¹² shots. It is overtaken by both Gemini models at higher shot counts. **6. Bottom-Right: English → Navajo** * **Y-Axis Range:** ~10.0 to 35.0. * **Trends & Key Points:** * **Gemini 1.5 Pro (Green):** Starts at ~24.5 (2⁰ shots). Shows a gradual increase until 2⁸ shots, then a sharper rise, ending at ~34.0 at 2¹² shots. * **Gemini 1.5 Flash (Blue):** Starts at ~11.0 (2⁰ shots). Dips slightly to ~10.0 at 2⁵ shots, then begins a strong, accelerating upward trend, ending at ~28.0 at 2¹² shots. * **GPT-4 Turbo (Gray):** Starts at ~18.0 (2⁰ shots). Shows a slow, steady increase, ending at ~23.0 at 2¹² shots. ### Key Observations 1. **Model Hierarchy:** Gemini 1.5 Pro is the top or tied-for-top performer in all six translation tasks, demonstrating consistently high baseline performance and scaling. 2. **Scaling Behavior:** Gemini 1.5 Flash shows the most dramatic scaling with increased shots, often starting lowest but exhibiting the steepest improvement curves (especially for Bemba, Acholi, Abkhaz, Navajo). Its performance is highly sensitive to the number of shots. 3. **GPT-4 Turbo's Plateau:** GPT-4 Turbo often shows the least improvement with more shots (e.g., Kurdish, Ewe, Acholi). Its performance lines are notably flatter, suggesting it may benefit less from additional in-context examples for these specific low-resource language tasks within the tested range. 4. **Language Difficulty:** The absolute chrF scores vary widely by language. For example, scores for Kurdish are high (35-47 range), while scores for Abkhaz and Navajo start much lower (5-35 range), indicating these are more challenging translation tasks for the models. 5. **Anomaly - Acholi Dip:** Gemini 1.5 Flash shows a distinct performance dip at 2⁵ (32) shots for English→Acholi before recovering, which is not observed in the other languages or models. ### Interpretation This data suggests a significant difference in how these advanced AI models utilize in-context learning for machine translation, particularly for lower-resource languages. * **Gemini 1.5 Pro** appears to have a robust underlying translation capability that requires less "priming" with examples, as evidenced by its high starting points. It still benefits from more shots, but the marginal gain is smaller. * **Gemini 1.5 Flash** behaves like a model that is highly reliant on in-context learning. Its poor zero/few-shot performance but strong scaling indicates it effectively uses the provided examples to adapt and improve, making it potentially more flexible but dependent on having a sufficient demonstration set. * **GPT-4 Turbo's** flat scaling for several languages could imply a few possibilities: its pre-training data may have included less relevant material for these languages, its context window utilization for this task is less efficient, or it has reached a performance plateau that more examples cannot easily break through. The stark contrast in scaling behaviors (steep vs. flat lines) highlights that "more shots" is not universally beneficial to the same degree. The optimal model choice may depend on the available computational budget for inference (more shots = more cost) and the specific target language. The charts provide a clear visual argument that model performance on low-resource language translation is not static but dynamically changes with the amount of provided context, and this dynamic varies fundamentally between model architectures.

## Line Graphs: Test ChrF (Floores/In-house) vs. Number of Shots (K) for Translation Tasks ### Overview The image contains six line graphs comparing the performance of three AI models (Gemini 1.5 Pro, Gemini 1.5 Flash, GPT-4 Turbo) across six translation tasks (English → Bemba, Kurdish, Ewe, Acholi, Abkhaz, Navajo). Each graph plots **Test ChrF** (a metric for translation quality) against **Number of Shots (K)**, where K represents the amount of training data used (e.g., 2^1 = 2 examples, 2^12 = 4096 examples). The graphs are arranged in two rows of three, with legends at the top indicating model colors: yellow (Gemini 1.5 Pro), blue (Gemini 1.5 Flash), gray (GPT-4 Turbo). --- ### Components/Axes - **X-axis**: Number of Shots (K) on a logarithmic scale (2^1 to 2^12). - **Y-axis**: Test ChrF (Floores/In-house), with values ranging from ~10 to 50 depending on the task. - **Legends**: Positioned at the top of all graphs, with consistent color coding: - Yellow: Gemini 1.5 Pro - Blue: Gemini 1.5 Flash - Gray: GPT-4 Turbo - **Subplot Titles**: Each graph is labeled with the translation direction (e.g., "Translation: English → Bemba"). --- ### Detailed Analysis #### 1. Translation: English → Bemba - **Trend**: Gemini 1.5 Pro (yellow) shows a steep upward slope, reaching ~50 Floores at 2^12 shots. Gemini 1.5 Flash (blue) increases from ~30 to ~45, while GPT-4 Turbo (gray) remains flat (~30–35). - **Key Data Points**: - At 2^1 shots: Gemini 1.5 Pro ≈ 40, Gemini 1.5 Flash ≈ 30, GPT-4 Turbo ≈ 30. - At 2^12 shots: Gemini 1.5 Pro ≈ 50, Gemini 1.5 Flash ≈ 45, GPT-4 Turbo ≈ 35. #### 2. Translation: English → Kurdish - **Trend**: Gemini 1.5 Pro (yellow) rises from ~40 to ~45, Gemini 1.5 Flash (blue) increases from ~35 to ~42, and GPT-4 Turbo (gray) stays flat (~30–35). - **Key Data Points**: - At 2^1 shots: Gemini 1.5 Pro ≈ 40, Gemini 1.5 Flash ≈ 35, GPT-4 Turbo ≈ 30. - At 2^12 shots: Gemini 1.5 Pro ≈ 45, Gemini 1.5 Flash ≈ 42, GPT-4 Turbo ≈ 35. #### 3. Translation: English → Ewe - **Trend**: Gemini 1.5 Pro (yellow) increases from ~40 to ~45, Gemini 1.5 Flash (blue) rises from ~30 to ~38, and GPT-4 Turbo (gray) remains flat (~25–30). - **Key Data Points**: - At 2^1 shots: Gemini 1.5 Pro ≈ 40, Gemini 1.5 Flash ≈ 30, GPT-4 Turbo ≈ 25. - At 2^12 shots: Gemini 1.5 Pro ≈ 45, Gemini 1.5 Flash ≈ 38, GPT-4 Turbo ≈ 30. #### 4. Translation: English → Acholi - **Trend**: Gemini 1.5 Pro (yellow) increases from ~30 to ~35, Gemini 1.5 Flash (blue) rises from ~15 to ~30, and GPT-4 Turbo (gray) stays flat (~20–25). - **Key Data Points**: - At 2^1 shots: Gemini 1.5 Pro ≈ 30, Gemini 1.5 Flash ≈ 15, GPT-4 Turbo ≈ 20. - At 2^12 shots: Gemini 1.5 Pro ≈ 35, Gemini 1.5 Flash ≈ 30, GPT-4 Turbo ≈ 25. #### 5. Translation: English → Abkhaz - **Trend**: Gemini 1.5 Pro (yellow) increases from ~25 to ~35, Gemini 1.5 Flash (blue) rises from ~10 to ~30, and GPT-4 Turbo (gray) remains flat (~20–25). - **Key Data Points**: - At 2^1 shots: Gemini 1.5 Pro ≈ 25, Gemini 1.5 Flash ≈ 10, GPT-4 Turbo ≈ 20. - At 2^12 shots: Gemini 1.5 Pro ≈ 35, Gemini 1.5 Flash ≈ 30, GPT-4 Turbo ≈ 25. #### 6. Translation: English → Navajo - **Trend**: Gemini 1.5 Pro (yellow) increases from ~25 to ~35, Gemini 1.5 Flash (blue) rises from ~10 to ~30, and GPT-4 Turbo (gray) stays flat (~20–25). - **Key Data Points**: - At 2^1 shots: Gemini 1.5 Pro ≈ 25, Gemini 1.5 Flash ≈ 10, GPT-4 Turbo ≈ 20. - At 2^12 shots: Gemini 1.5 Pro ≈ 35, Gemini 1.5 Flash ≈ 30, GPT-4 Turbo ≈ 25. --- ### Key Observations 1. **Model Performance**: - **Gemini 1.5 Pro** consistently outperforms other models across all tasks, with the largest gains at higher shot counts (e.g., +10 Floores from 2^1 to 2^12 in Bemba). - **Gemini 1.5 Flash** shows moderate improvement, with gains of ~10–15 Floores across tasks. - **GPT-4 Turbo** performs the least effectively, with minimal improvement (often flat lines). 2. **Task-Specific Trends**: - **Low-Resource Languages** (e.g., Abkhaz, Navajo): Gemini 1.5 Pro and Flash show steeper improvement curves, suggesting better scalability with limited data. - **High-Resource Languages** (e.g., Kurdish, Ewe): GPT-4 Turbo maintains higher baseline performance but fails to scale. 3. **Anomalies**: - In **English → Acholi**, Gemini 1.5 Flash shows a sharp dip at 2^3 shots (~12 Floores) before recovering, possibly indicating overfitting or data noise. --- ### Interpretation The data demonstrates that **Gemini 1.5 Pro** excels in translation quality, particularly when trained on larger datasets (higher K values). Its performance scales nearly linearly with the number of shots, suggesting robust generalization. **Gemini 1.5 Flash** also improves with more data but lags behind the Pro version, likely due to architectural differences (e.g., smaller model size). **GPT-4 Turbo** performs poorly in low-resource scenarios (e.g., Abkhaz, Navajo) but maintains mediocre results in high-resource tasks, indicating limited adaptability to new languages. The stark contrast between Gemini models and GPT-4 Turbo highlights the importance of model architecture and training strategies for cross-lingual tasks. The dip in Gemini 1.5 Flash for Acholi warrants further investigation into data quality or model stability. Overall, these results underscore the value of iterative training with increasing data volumes for low-resource language translation.