Datasets:
source stringclasses 1
value | model stringclasses 1
value | model_requested stringclasses 1
value | model_canonical stringclasses 1
value | model_name stringclasses 1
value | served_model stringclasses 1
value | messages listlengths 3 3 | id int64 0 249 | category stringclasses 7
values | usage dict | finish_reason stringclasses 1
value | latency_s float64 2.74 640 |
|---|---|---|---|---|---|---|---|---|---|---|---|
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "What defines consciousness, and can artificial intelligence ever truly possess it?"
},
{
"role": "assistant",
"content": "## First, disentangle th... | 1 | 9_humanities | {
"prompt_tokens": 54,
"completion_tokens": 2569,
"total_tokens": 2623,
"cost": 0.0116415,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.0116415,
"up... | stop | 54.3 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "What role should diagramming and visualization play in making requirements understandable across disciplines?"
},
{
"role": "assistant",
"content"... | 3 | 2_cs | {
"prompt_tokens": 56,
"completion_tokens": 2688,
"total_tokens": 2744,
"cost": 0.01218,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.01218,
"upstre... | stop | 57.75 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "How might requirements engineering benefit from knowledge graphs, semantic web technologies, or ontologies?"
},
{
"role": "assistant",
"content": ... | 4 | 2_cs | {
"prompt_tokens": 58,
"completion_tokens": 3614,
"total_tokens": 3672,
"cost": 0.01635,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.01635,
"upstre... | stop | 73.51 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "What are the risks and opportunities of using generative AI to assist with requirements elicitation, analysis, or documentation?"
},
{
"role": "assist... | 5 | 2_cs | {
"prompt_tokens": 64,
"completion_tokens": 4159,
"total_tokens": 4223,
"cost": 0.0188115,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.0188115,
"up... | stop | 85.95 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "Write Python code to implement a neural network from scratch without libraries."
},
{
"role": "assistant",
"content": "# Neural network from scrat... | 2 | 1_code | {
"prompt_tokens": 54,
"completion_tokens": 10996,
"total_tokens": 11050,
"cost": 0.049563,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.049563,
"up... | stop | 167.78 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "How can we handle requirements from stakeholders who themselves don’t fully understand the problem or solution space?"
},
{
"role": "assistant",
"... | 6 | 2_cs | {
"prompt_tokens": 60,
"completion_tokens": 2718,
"total_tokens": 2778,
"cost": 0.012321,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.012321,
"upst... | stop | 60.37 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "What strategies can a requirements engineer use when stakeholders change their minds frequently or inconsistently?"
},
{
"role": "assistant",
"con... | 7 | 2_cs | {
"prompt_tokens": 58,
"completion_tokens": 3306,
"total_tokens": 3364,
"cost": 0.014964,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.014964,
"upst... | stop | 73.29 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "What can we do when the problem itself is unclear, and requirements are being defined before the problem is fully understood?"
},
{
"role": "assistant... | 8 | 2_cs | {
"prompt_tokens": 64,
"completion_tokens": 2984,
"total_tokens": 3048,
"cost": 0.013524,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.013524,
"upst... | stop | 58.58 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "How might we prioritize requirements in a system where competing interests are equally valid?"
},
{
"role": "assistant",
"content": "Equal validit... | 9 | 2_cs | {
"prompt_tokens": 56,
"completion_tokens": 2675,
"total_tokens": 2731,
"cost": 0.0121215,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.0121215,
"up... | stop | 56.42 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "How might we recover requirements from a failed or poorly documented project?"
},
{
"role": "assistant",
"content": "## First, decide what \"recov... | 10 | 2_cs | {
"prompt_tokens": 54,
"completion_tokens": 3179,
"total_tokens": 3233,
"cost": 0.0143865,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.0143865,
"up... | stop | 63.59 |
TeichAI/claude-4.5-opus-high-reasoning-250x | qwen/qwen4-max | qwen/qwen4-max | qwen/qwen4-max-20261115 | Qwen: Qwen4 Max | qwen/qwen4-max | [
{
"role": "system",
"content": "You're an expert at reasoning, and give thoughtful and precise answers."
},
{
"role": "user",
"content": "Can requirements ever be “complete,” or is completeness an illusion? Why or why not?"
},
{
"role": "assistant",
"content": "## Short answer\n\nAbs... | 12 | 2_cs | {
"prompt_tokens": 59,
"completion_tokens": 2459,
"total_tokens": 2518,
"cost": 0.011154,
"is_byok": false,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.011154,
"upst... | stop | 49.24 |
Qwen4 Max Reasoning Traces — 250x
250 prompts from the MiMo-2.6-Pro 250x prompt set, answered by qwen/qwen4-max
(OpenRouter, canonical qwen/qwen4-max-20261115) on 2026-11-15/16. One row per prompt,
with the full reasoning trace and final answer.
| metric | value |
|---|---|
| rows | 250 |
| output tokens | 1,672,656 |
| input tokens | 17,517 |
| reasoning chars | 2,616,051 |
| content chars | 3,164,330 |
| output tokens / row | 6,691 mean, 3,706 median, 43,473 max |
| retained cost | $7.5532 ($0.0302/row) |
| total billed | $7.9960 |
| wall time | 2h 22m 50s |
| finish reasons | 250 stop, 0 truncated, 0 failed |
Files
| file | contents |
|---|---|
corpus.jsonl |
250 records: prompt + Qwen4 Max reasoning + answer |
stats.json |
row/category/token/cost totals and run settings |
Schema
Each line is one JSON object:
| field | contents |
|---|---|
messages |
system, user, and assistant; the assistant message carries reasoning and content |
id |
original 0-based row index in the MiMo 250x set |
category |
local judge label (1_code, 2_cs, 4_stats, 5_finance, 6_business, 8_science, 9_humanities), or null for the 105 uncategorized rows |
source |
upstream prompt dataset recorded on the row (TeichAI/claude-4.5-opus-high-reasoning-250x) |
model |
requested model id (qwen/qwen4-max) |
model_canonical |
dated variant the alias resolved to (qwen/qwen4-max-20261115) |
model_name, served_model |
provider metadata |
usage |
OpenRouter usage for the row, including prompt_tokens, completion_tokens, and cost |
finish_reason |
stop for all 250 rows |
latency_s |
wall time of the request |
The system prompt is identical on every row; id preserves the original prompt index so
subsets can be joined to the MiMo/TeichAI 250x sets.
Categories
Categories are local judge labels, not fields in either upstream dataset. The 105 uncategorized rows are included as-is; they are a heterogeneous mix, not a separate task.
| category | rows | output tokens | cost |
|---|---|---|---|
1_code |
32 | 322,404 | $1.4537 |
uncategorized |
105 | 307,475 | $1.3946 |
5_finance |
13 | 247,979 | $1.1175 |
6_business |
22 | 205,425 | $0.9270 |
2_cs |
29 | 201,980 | $0.9118 |
9_humanities |
32 | 160,886 | $0.7272 |
4_stats |
9 | 136,147 | $0.6137 |
8_science |
8 | 90,360 | $0.4076 |
Generation
- Teacher:
qwen/qwen4-maxvia OpenRouter,reasoning: enabled, streaming. - Sampling: temperature 0.7, no top-p override.
- First pass:
max_tokens=32768, concurrency 4, 7,928 s. - Three rows (ids 25, 30, 95) hit the 32,768 cap and were regenerated at
max_tokens=65536; all three then finished withstop(641 s, $0.3902). - Total billed includes the discarded truncated attempts;
stats.jsonreports bothcost_usd_retainedandcost_usd_billed.
Provenance
- Prompts:
TeichAI/claude-4.5-opus-high-reasoning-250x(Claude Opus 4.5 high-reasoning prompt set), reused byxtools-at/Xiaomi-MiMo-2.6-Pro-High-Reasoning-250x. - Responses: Qwen4 Max, generated for this dataset; not copied from either upstream set.
- Categories and profiles: local judge-labelled splits of the same 250 prompts.
- Generator:
scripts/openrouter_reasoning_corpus.pyfrom the LoNE experiment repository; the standalone gist is https://gist.github.com/AJ-Gonzalez/31f1f8cfe45c233911e055d5cfed6da0
Usage
from datasets import load_dataset
ds = load_dataset("AJ-Gonzalez/qwen4-max-reasoning-250x", split="train")
row = ds[0]
print(row["messages"][-1]["reasoning"][:500])
print(row["messages"][-1]["content"][:500])
Filter the coder profile (the same definition used by the generator):
coder = ds.filter(lambda r: r["category"] in {"1_code", "2_cs", "4_stats"})
Limitations
- Model outputs, including the reasoning text, are not verified for correctness. They may contain factual errors, flawed reasoning, or unsafe code.
- The reasoning field is generated text, not a verified record of the model's internal computation.
- The 105 uncategorized rows vary widely in length (46 to 43,473 output tokens); they are not a balanced evaluation set.
- These prompts have been answered by several other models. This dataset is one teacher's responses to the same inputs, not a new benchmark.
Licence
license: other. The upstream prompt dataset (TeichAI) and the MiMo set that redistributes
those prompts both use license: other; the MiMo card asks users to review the upstream
terms and the rights applicable to generated outputs before redistribution or commercial
use. Qwen4 Max was accessed through OpenRouter, so the Qwen and OpenRouter terms also apply.
No new licence is asserted over the model-generated traces here. Review all upstream terms
before commercial use.
- Downloads last month
- 296