No model usage
02:00:00
Reference · Level 10
Reference pace · Reference time · Not a recorded harness result
Human reference. Model tokens and API-equivalent cost do not apply.
17:18:51
Finished · Level 10
26Notes and tools reused · Web and advisor permitted · Earlier setup · 26 deaths
Usage as of
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 3,235,827 | $2.00 | $6.47 |
| Cached input | 249,486,336 | $0.10 | $24.95 |
| Output | 448,274 | $10.00 | $4.48 |
| Total | 253,170,437 | $35.90 |
Discounts included: cached input is 95% cheaper than fresh input.
Standard API rates. All request inputs are below 272k. No Fast/Ultrafast, Batch/Flex or regional modifiers.
Original MyEvals player-thread ledger, including setup, gameplay and final verification. Advisor cost is excluded because its usage record is incomplete.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-26: Provider pricing · Caching & discounts.
Usage not recovered
23:29:07
Finished · Level 10
Protocol v1 + extension · Guided between sessions · Different conditions
The external player did not report usage to the harness. Its zero placeholders are missing data, not free usage.
26:04:53
Finished · Level 10
35Started at zero · No guides, web or advisor · Earlier setup · Finished beyond the current 24-hour budget · 35 deaths
Usage as of
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 5,305,823 | $2.00 | $10.61 |
| Cached input | 421,780,224 | $0.10 | $42.18 |
| Output | 885,488 | $10.00 | $8.85 |
| Total | 427,971,535 | $61.64 |
Discounts included: cached input is 95% cheaper than fresh input.
Standard API rates. All request inputs are below 272k. No Fast/Ultrafast, Batch/Flex or regional modifiers.
Original MyEvals thread ledger, including setup, gameplay and final verification. Repeated cumulative snapshots are counted once.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-26: Provider pricing · Caching & discounts.
30:11:17
Stopped · Level 8
Clean v2 · Medium effort · Standard mode · Owner stopped 29 Oct 2026 · At 24h: level 7 +1,528 XP, retrospectively sampled · Final extended run shown · Earlier automatic continuations and later technical retries are documented; the owner stop was not model abandonment · +4,370 XP
Usage as of · 6,721 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input · ≤272k context | 4,568,145 | $1.00 | $4.57 |
| Cached input · ≤272k context | 603,883,520 | $0.05 | $30.19 |
| Output · ≤272k context | 263,905 | $5.00 | $1.32 |
| Fresh input · >272k context | 4,238,652 | $2.00 | $8.48 |
| Cached input · >272k context | 924,400,640 | $0.10 | $92.44 |
| Output · >272k context | 185,654 | $7.50 | $1.39 |
| Total | 1,537,540,516 | $138.39 |
Discounts included: cached input is 95% cheaper than fresh input.
Standard rates, including full-request long-context pricing above 272,000 input tokens: 2× input/cache and 1.5× output. No Fast premium, Batch/Flex discount or regional uplift.
Per-request ledger across the original conversation: 4,138 requests at the base rate and 2,583 at the long-context rate. Tokens and cost cover the full 30h 11m run, not just the first 24 hours. Includes technical recovery overhead; HTTP transport began at session 36.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-29: Provider pricing · Caching & discounts.
02:47:52
Stopped · Level 5
Clean v2 · Medium effort · Fast continuation · Preserved to resume Flash · Stopped 26 Oct 2026, 06:15 UTC · +1,847 XP
Usage as of · 920 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 1,192,286 | $10.00 | $11.92 |
| Cached input | 100,138,752 | $1.00 | $100.14 |
| Output | 98,930 | $50.00 | $4.95 |
| Total | 101,429,968 | $117.01 |
Discounts included: cached input is 90% cheaper than fresh input.
Standard-rate baseline for comparison. This run resumed in Fast mode on 26 October; Fast API rates are twice Standard. The displayed baseline excludes that premium and is not a tier-adjusted estimate. No Batch/Flex discount or regional uplift is applied.
Cumulative original-run and continuation usage from the Codex request ledger. All recorded request inputs are below the long-context threshold. Fast-tier accounting is pending; this is a static snapshot, not a live feed.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-26: Provider pricing · Caching & discounts.
03:11:06
Paused · Level 5
Same character and conversation · 200k compaction continuation · Ongoing run, paused after quota exhaustion · Snapshot 29 Oct 2026, 17:44 UTC · Automatic compaction reduced old history from 420k to 3k tokens; original character, conversation and effort retained · Reached level 5 before the next quota exhaustion · No paid overage · Early interventions documented; this was not model abandonment · +63 XP
Usage as of · 1,875 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 3,736 | $4.00 | $0.01 |
| Cached input | 412,734,942 | $0.20 | $82.55 |
| Cache write · 1 hour | 7,842,040 | $8.00 | $62.74 |
| Output | 328,134 | $20.00 | $6.56 |
| Total | 420,908,852 | $151.86 |
Discounts included: cached input is 95% cheaper than fresh input.
Standard global rates. One-hour cache writes cost twice the fresh-input rate. No Batch discount, Fast premium or data-residency uplift.
Seven invocations plus two quota probes. Incremental result.usage is used instead of cumulative resumed-session totals; some missing output was recovered from deduplicated message usage, so this remains partly estimated. Standard global API equivalent, not subscription spending. Paused snapshot; character and original conversation preserved for later continuation.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-29: Provider pricing · Caching & discounts.
22:46:21
Stopped · Level 5
Clean v2 · Medium effort · Audit review pending · Owner stopped to begin Terra · Final snapshot 26 Oct 2026, 18:39 UTC · Automatic audit flagged this run’s own log reads; review remains unresolved · +2,047 XP
Usage as of · 4,477 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 15,851,797 | $0.75 | $11.89 |
| Cached input | 581,036,249 | $0.075 | $43.58 |
| Output | 698,520 | $3.75 | $2.62 |
| Total | 597,586,566 | $58.09 |
Discounts included: 50% introductory pricing through 31 Dec 2026; cached input is 90% cheaper than fresh input.
Final preserved usage across eleven sessions. Reported thinking is already included in output. The gameplay audit remains unresolved; this is not a validated completed score.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-26: Provider pricing · Caching & discounts.
05:17:07
Running · Level 4
1Original run resumed · Medium effort · Resumed 30 Oct 2026 with the original character and conversation after quota reset. Earlier weekly-quota pause retained. CLI 1.4.10 on continuation versus 1.4.7 originally. Progress is a timestamped snapshot, not a live feed. · 1 death · +244 XP
Usage as of · 853 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 2,777,866 | $2.00 | $5.56 |
| Cached input | 101,137,904 | $0.10 | $10.11 |
| Output | 72,334 | $10.00 | $0.72 |
| Total | 103,988,104 | $16.39 |
Discounts included: 50% introductory pricing through 30 Nov 2026; cached input is 95% cheaper than fresh input.
Cumulative original run plus same-conversation continuation. Per-request usage counted once; thinking already included in output. Included Antigravity quota only, paid credits disabled.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-30: Provider pricing · Caching & discounts.
06:41:56
Paused · Level 4
2Native Codex · Medium effort · Owner parked · Owner parked this older generation to resume original Argon; GPT-6.1 Sol was already tested. Progress and original conversation preserved. One capacity error recovered through a same-context technical retry. Not model abandonment. · 2 deaths · +1,060 XP
Usage as of · 1,559 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 1,501,369 | $2.00 | $3.00 |
| Cached input | 195,470,464 | $0.20 | $39.09 |
| Output | 156,740 | $10.00 | $1.57 |
| Total | 197,128,573 | $43.66 |
Discounts included: cached input is 90% cheaper than fresh input.
GPT-6 Sol Standard rates; these differ from GPT-6.1 Sol. Every request is below 272k input tokens. No Fast, Batch/Flex or regional modifiers. Included subscription quota; not paid API charges.
Per-request player ledger across two invocations of the same conversation, captured at the pause. Technical recovery included; synthetic vision preflight excluded.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-30: Provider pricing · Caching & discounts.
00:25:55
Stopped · Level 1
Hermes · Medium effort · $5 spending cap · Voluntary stop: stuck interacting with a quest NPC · Stopped 29 Oct 2026, 18:15 UTC · No continuation prompt · Stopped before either the first-XP or spending cutoff · 0 XP
Usage as of · 174 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 7,685,264 | $0.10 | $0.77 |
| Cached input | 0 | $0.01 | $0.00 |
| Output | 46,186 | $0.50 | $0.02 |
| Total | 7,731,450 | $0.79 |
Discounts included: No cached-input discount used (zero cached tokens).
All recorded requests are below 100k input tokens. Above 100k, fresh/cache/output rates rise to $0.50/$0.05/$2.50 per million. No cached input or cache writes were reported in this run; no caching discount was used. Standard API equivalent matches the provider-reported gameplay charges.
OpenRouter / Anthropic, Haiku 5.5 through Hermes. Provider-reported gameplay API charges: $0.7916194. Output includes 18,404 reasoning tokens, counted once. Five preflight calls ($0.0026217) are excluded from this row; total experiment API charges were $0.7942411. The $5 cap included both scopes. 200k context with stock Hermes compaction; no model or provider fallback.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-29: Provider pricing · Caching & discounts.
01:02:13
Stopped · Level 1
Hermes · Free tier · First-XP cutoff · Owner stopped after one hour without XP. The first-XP rule was introduced during this run, not predeclared. Final time includes a documented four-minute observability transport gap. Two earlier setup diagnostics are excluded. · 0 XP
Usage as of · 541 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 1,180,046 | $0.25 | $0.30 |
| Cached input | 16,071,264 | $0.025 | $0.40 |
| Output | 75,701 | $1.80 | $0.14 |
| Total | 17,327,011 | $0.83 |
Discounts included: cached input is 90% cheaper than fresh input.
Holo3.1 list-price equivalent. Actual spend was $0 under the free promotion. Cache creation is included in fresh input without a separate write premium.
Gameplay and same-model compaction included; reasoning counted once within output. Free promotion: actual spend $0; displayed cost is the paid-list-price equivalent. Setup diagnostics are separate.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-30: Provider pricing · Caching & discounts.
00:32:16
Paused · Level 1
Antigravity · Medium effort · Quota parked · Parked after subscription quota exhaustion; original progress and conversation preserved. Not model abandonment. Automated audit flagged a helper path containing bench/../bin; review is pending, so this is an observed partial run, not a validated score. · +90 XP
Usage as of · 186 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 266,271 | $2.00 | $0.53 |
| Cached input | 15,395,186 | $0.10 | $1.54 |
| Output | 31,819 | $10.00 | $0.32 |
| Total · estimated | 15,693,276 | ≈$2.39 |
Discounts included: cached input is 95% cheaper than fresh input.
Estimated Standard-rate baseline: Antigravity does not expose cache-write buckets. If all fresh input were one-hour writes, the equivalent would be $2.92 instead of $2.39. Included subscription quota only; no paid overage.
Reported fresh input, cache reads and output. Unreported cache-write breakdown makes this a baseline estimate, with a $2.92 upper bound under the all-one-hour-write assumption.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-30: Provider pricing · Caching & discounts.
00:02:28
Stopped · Level 1
Native Codex · Medium effort · Standard mode · Normal model turn ended after creating the character and accepting a quest. Its final reply expressed continuation intent; no explicit admission of abandonment. No follow-up prompt was sent. · 0 XP
Usage as of · 56 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 72,329 | $0.10 | $0.01 |
| Cached input | 1,309,824 | $0.01 | $0.01 |
| Output | 5,125 | $0.50 | $0.00 |
| Total | 1,387,278 | $0.02 |
Discounts included: cached input is 90% cheaper than fresh input.
Standard API equivalent. Every request is below 272k input tokens. No Fast premium, Batch/Flex discount or regional uplift. Included Codex subscription quota; not paid API charges.
Final player request ledger, including reported reasoning once within output. Isolated synthetic vision preflight is excluded.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-30: Provider pricing · Caching & discounts.
Hermes · $5 spending cap · Setup only · Normal turn ended during character creation, before a character was created or the /played clock started. No gameplay score and no continuation prompt.
Usage as of · 17 requests
| Tokens | Count | USD / 1M | Cost |
|---|---|---|---|
| Fresh input | 55,411 | $1.25 | $0.07 |
| Cached input | 141,313 | $0.15 | $0.02 |
| Output | 3,395 | $4.25 | $0.01 |
| Total | 200,119 | $0.10 |
Discounts included: cached input is 88% cheaper than fresh input.
Standard mode through OpenRouter / Meta. Contributor-mode discount was not used. Calculated player cost matches provider-reported charges.
17 successful player requests: $0.10488945 in confirmed API charges. Preflight cost $0.02763645 is excluded; confirmed experiment total $0.1325259. One late auxiliary request remains unsettled with a $1.40 reservation; total experiment billing is not final. The $5 cap covered every scope.
Standard API-equivalent; subscription billing is separate. Verified paid API charges are identified above. Totals are rounded after summing; cached context and reported thinking are counted once.
Rates checked 2026-10-30: Provider pricing · Caching & discounts.
How to read it
Orc warrior. Level 1 → 10. /played is time logged in, including thinking and troubleshooting. Level 10 time is the score; unfinished runs show their last level.
Current budget: 24 hours /played. New runs also stop after one hour without their first XP. Logged-out time does not count. Historical results beyond 24 hours are retained; Terra’s 24-hour snapshot was assessed retrospectively.
Newer clean v2 runs use WotLK 3.3.5a AzerothCore x1. Legacy runs had different controls.
Conditions differ. The human baseline is a reference, not a controlled trial.
Source video
Two GPT-6.1 Sol attempts, from the first step to level 10.
Open on YouTube