Session Date: January 20, 2026 Analysis Period: December 9, 2025 - January 20, 2026 Data Source: ~/.claude/stats-cache.json, git history
Executive Summary
- Total Cost: $3,674.37 over 42 active days
- Total Sessions: 306 sessions
- Total Messages: 110,288 messages
- Key Finding: 69% reduction in cost-per-message from December to January
- Implementation: Added context window tracking to session-start hooks
1. January 2026 Daily Activity
| Date | Messages | Sessions | Tool Calls | Msgs/Session |
|---|
| Jan 3 | 1,569 | 3 | 469 | 523 |
| Jan 8 | 100 | 1 | 34 | 100 |
| Jan 10 | 6,627 | 31 | 1,891 | 214 |
| Jan 11 | 9,022 | 22 | 2,629 | 410 |
| Jan 12 | 30 | 1 | 10 | 30 |
| Jan 14 | 501 | 3 | 134 | 167 |
| Jan 16 | 7,543 | 16 | 1,049 | 471 |
| Jan 17 | 22,124 | 16 | 1,420 | 1,383 |
| Jan 18 | 6,019 | 11 | 573 | 547 |
| Jan 19 | 19,123 | 83 | 2,902 | 230 |
| Jan 20 | 8,721 | 31 | 1,352 | 281 |
January Token Usage by Model
| Date | Opus 4.5 | Haiku 4.5 | Sonnet 4.5 | Total |
|---|
| Jan 3 | 67,376 | 0 | 0 | 67,376 |
| Jan 8 | 20,377 | 0 | 0 | 20,377 |
| Jan 10 | 590,831 | 12,023 | 0 | 602,854 |
| Jan 11 | 1,315,439 | 12,901 | 47,379 | 1,375,719 |
| Jan 12 | 5,207 | 0 | 0 | 5,207 |
| Jan 14 | 29,742 | 1,243 | 0 | 30,985 |
| Jan 16 | 149,389 | 1,598 | 312 | 151,299 |
| Jan 17 | 97,184 | 211 | 6,871 | 104,266 |
| Jan 18 | 85,665 | 320 | 0 | 85,985 |
| Jan 19 | 497,342 | 1,302 | 49 | 498,693 |
| Jan 20 | 180,740 | 0 | 0 | 180,740 |
January Weekly Comparison
| Period | Messages | Sessions | Tokens | Avg Tokens/Session |
|---|
| Week 1 (Jan 1-7) | 1,569 | 3 | 67,376 | 22,459 |
| Week 2 (Jan 8-14) | 16,280 | 58 | 2,015,142 | 34,744 |
| Week 3 (Jan 15-20) | 63,530 | 157 | 1,020,983 | 6,503 |
2. December 2025 Daily Activity
| Date | Messages | Sessions | Tool Calls | Msgs/Session |
|---|
| Dec 9 | 1,852 | 5 | 651 | 370 |
| Dec 10 | 1,714 | 10 | 625 | 171 |
| Dec 11 | 1,574 | 2 | 572 | 787 |
| Dec 12 | 127 | 1 | 44 | 127 |
| Dec 14 | 9,572 | 27 | 2,996 | 355 |
| Dec 16 | 577 | 5 | 159 | 115 |
| Dec 20 | 757 | 1 | 218 | 757 |
| Dec 21 | 115 | 1 | 35 | 115 |
| Dec 24 | 5,585 | 14 | 1,714 | 399 |
| Dec 25 | 1,774 | 7 | 517 | 253 |
| Dec 26 | 200 | 2 | 50 | 100 |
| Dec 27 | 5,062 | 13 | 1,401 | 389 |
December Token Usage by Model
| Date | Opus 4.5 | Haiku 4.5 | Sonnet 4.5 | Total |
|---|
| Dec 9 | 180,838 | 7,604 | 4,390 | 192,832 |
| Dec 10 | 142,945 | 12,550 | 0 | 155,495 |
| Dec 11 | 166,471 | 9,583 | 0 | 176,054 |
| Dec 12 | 77,116 | 0 | 0 | 77,116 |
| Dec 14 | 1,055,057 | 22,840 | 0 | 1,077,897 |
| Dec 16 | 146,036 | 2,011 | 0 | 148,047 |
| Dec 20 | 107,681 | 0 | 0 | 107,681 |
| Dec 21 | 8,470 | 0 | 0 | 8,470 |
| Dec 24 | 823,447 | 11,348 | 0 | 834,795 |
| Dec 25 | 301,135 | 7,689 | 0 | 308,824 |
| Dec 26 | 5,466 | 0 | 0 | 5,466 |
| Dec 27 | 443,140 | 5,816 | 0 | 448,956 |
3. Monthly Comparison: December vs January
Totals Comparison
| Metric | December | January | Change |
|---|
| Messages | 28,909 | 81,379 | +181% |
| Sessions | 88 | 218 | +148% |
| Tool Calls | 8,982 | 12,463 | +39% |
| Total Tokens | 3,541,633 | 3,123,501 | -12% |
| Active Days | 12 | 11 | -8% |
Averages Comparison
| Metric | December | January | Change |
|---|
| Msgs/Day | 2,409 | 7,398 | +207% |
| Sessions/Day | 7.3 | 19.8 | +171% |
| Tools/Day | 749 | 1,133 | +51% |
| Tokens/Day | 295,136 | 283,955 | -4% |
| Tokens/Session | 40,246 | 14,328 | -64% |
| Tokens/Message | 122.5 | 38.4 | -69% |
Model Distribution
| Model | December | January | Change |
|---|
| Opus 4.5 | 3,457,802 (97.6%) | 3,039,292 (97.3%) | -12% |
| Haiku 4.5 | 79,441 (2.2%) | 29,598 (0.9%) | -63% |
| Sonnet 4.5 | 4,390 (0.1%) | 54,611 (1.7%) | +1144% |
4. Cost Analysis
API Pricing (per million tokens)
| Model | Input | Output | Cache Read (0.1x) | Cache Write (1.25x) |
|---|
| Opus 4.5 | $5.00 | $25.00 | $0.50 | $6.25 |
| Sonnet 4.5 | $3.00 | $15.00 | $0.30 | $3.75 |
| Haiku 4.5 | $1.00 | $5.00 | $0.10 | $1.25 |
All-Time Token Usage
| Model | Input | Output | Cache Read | Cache Write |
|---|
| Opus 4.5 | 2.76M | 3.74M | 3,806M | 256.2M |
| Sonnet 4.5 | 42.7K | 16.3K | 11.4M | 3.2M |
| Haiku 4.5 | 32.1K | 77.0K | 102.9M | 29.0M |
Cost Breakdown by Model
Claude Opus 4.5
| Category | Tokens | Rate | Cost | |———-|——–|——|——| | Input | 2,755,560 | $5.00/M | $13.78 | | Output | 3,741,534 | $25.00/M | $93.54 | | Cache Read | 3,805,996,551 | $0.50/M | $1,903.00 | | Cache Write | 256,194,956 | $6.25/M | $1,601.22 | | Subtotal | | | $3,611.54 |
Claude Sonnet 4.5
| Category | Tokens | Rate | Cost | |———-|——–|——|——| | Input | 42,689 | $3.00/M | $0.13 | | Output | 16,312 | $15.00/M | $0.24 | | Cache Read | 11,362,461 | $0.30/M | $3.41 | | Cache Write | 3,240,361 | $3.75/M | $12.15 | | Subtotal | | | $15.93 |
Claude Haiku 4.5
| Category | Tokens | Rate | Cost | |———-|——–|——|——| | Input | 32,050 | $1.00/M | $0.03 | | Output | 76,989 | $5.00/M | $0.38 | | Cache Read | 102,914,527 | $0.10/M | $10.29 | | Cache Write | 28,960,773 | $1.25/M | $36.20 | | Subtotal | | | $46.90 |
Total Cost Summary
| Model | Cost | % of Total |
|---|
| Opus 4.5 | $3,611.54 | 98.3% |
| Haiku 4.5 | $46.90 | 1.3% |
| Sonnet 4.5 | $15.93 | 0.4% |
| TOTAL | $3,674.37 | 100% |
Cost by Category
| Category | Cost | % of Total |
|---|
| Cache Write | $1,649.57 | 44.9% |
| Cache Read | $1,916.70 | 52.2% |
| Output | $94.16 | 2.6% |
| Input | $13.94 | 0.4% |
Monthly Cost Breakdown
| Period | Token Share | Estimated Cost |
|---|
| December 2025 | 53.1% | $1,951.10 |
| January 2026 | 46.9% | $1,723.27 |
| Total | 100% | $3,674.37 |
Cost Efficiency Metrics
| Metric | December | January | Change |
|---|
| Total Cost | $1,951.10 | $1,723.27 | -12% |
| Daily Avg Cost | $162.59 | $156.66 | -4% |
| Cost/Session | $22.17 | $7.90 | -64% |
| Cost/Message | $0.067 | $0.021 | -69% |
| Cost/Tool Call | $0.217 | $0.138 | -36% |
5. Spike Day Analysis
All Spike Days Comparison
| Day | Type | Sessions | Tokens | Tokens/Session | Commits |
|---|
| Dec 14 | Research | 27 | 1.08M | 39,922 | 0 |
| Dec 24 | Research | 14 | 835K | 59,628 | 0 |
| Jan 11 | Research | 22 | 1.38M | 62,533 | 0 |
| Jan 17 | Implementation | 16 | 104K | 6,517 | 13 |
| Jan 19 | Rapid iteration | 83 | 499K | 6,008 | 12 |
December 14 Analysis
The Numbers: | Metric | Dec 14 | Dec Avg | vs Avg | |——–|——–|———|——–| | Tokens | 1,077,897 | 295,136 | 3.7x | | Messages | 9,572 | 2,409 | 4.0x | | Sessions | 27 | 7.3 | 3.7x | | Tool Calls | 2,996 | 749 | 4.0x |
Cause: Research sprint for OpenTelemetry integration (shipped Dec 27)
December 24 Analysis
The Numbers: | Metric | Dec 24 | Dec Avg | vs Avg | |——–|——–|———|——–| | Tokens | 834,795 | 295,136 | 2.8x | | Messages | 5,585 | 2,409 | 2.3x | | Sessions | 14 | 7.3 | 1.9x | | Tool Calls | 1,714 | 749 | 2.3x |
Cause: Final research push before Dec 27 OTel implementation
January 11 Analysis
The Numbers: | Metric | Jan 11 | Average | vs Avg | |——–|——–|———|——–| | Tokens | 1,375,719 | 283,955 | 4.8x | | Messages | 9,022 | 7,398 | 1.2x | | Sessions | 22 | 19.8 | 1.1x | | Tool Calls | 2,629 | 1,133 | 2.3x |
Cause: Research for SigNoz integration (shipped Jan 16)
January 17 Analysis — Anomaly Day
The Numbers: | Metric | Jan 17 | Jan Avg | vs Avg | |——–|——–|———|——–| | Messages | 22,124 | 7,398 | 3.0x (HIGHEST) | | Tokens | 104,266 | 283,955 | 0.37x (LOW) | | Sessions | 16 | 19.8 | 0.8x | | Tool Calls | 1,420 | 1,133 | 1.3x |
Unique Pattern: Highest messages, lowest tokens = Implementation day (not research)
Efficiency: 4.7 tokens/message vs 152.5 on Jan 11 (32x more efficient)
January 19 Analysis — Session Explosion
The Numbers: | Metric | Jan 19 | Jan Avg | vs Avg | |——–|——–|———|——–| | Messages | 19,123 | 7,398 | 2.6x | | Sessions | 83 | 19.8 | 4.2x (RECORD) | | Tool Calls | 2,902 | 1,133 | 2.6x | | Tokens | 498,693 | 283,955 | 1.8x |
Cause: High-velocity shipping day with frequent context resets
6. Research vs Implementation Pattern
Token Efficiency by Activity Type
| Type | Example | Messages | Tokens | Tokens/Msg |
|---|
| Research | Jan 11, Dec 14 | Medium | Very High | 100-150 |
| Implementation | Jan 17 | Very High | Low | 4-5 |
OTel Project Total (Dec 9 - Dec 28)
| Phase | Dates | Tokens | % of Project |
|---|
| Sprint 1 | Dec 9-16 | 1,826,321 | 48% |
| Break | Dec 17-23 | 116,151 | 3% |
| Sprint 2 | Dec 24-26 | 1,149,085 | 30% |
| Ship | Dec 27-28 | 723,269 | 19% |
| Total | | 3,814,826 | 100% |
SigNoz Project Total (Jan 10 - Jan 19)
| Phase | Dates | Tokens | Commits |
|---|
| Research | Jan 10-11 | 1.98M | 0 |
| Quiet | Jan 12-14 | 36K | 0 |
| Implementation | Jan 16 | 151K | 6 |
| Polish | Jan 17-19 | 689K | 25 |
7. Context Utilization Patterns
Cache Statistics
| Model | Cache Read | Cache Write | Read:Write Ratio |
|---|
| Opus 4.5 | 3,806M | 256M | 14.9:1 |
| Haiku 4.5 | 103M | 29M | 3.6:1 |
| Sonnet 4.5 | 11M | 3.2M | 3.5:1 |
Estimated Context Per Session
| Period | Sessions | Est. Context/Session |
|---|
| December | 88 | ~1.5M tokens |
| January | 218 | ~550K tokens |
| Change | +148% | -63% |
Context Trend by Day
| Day | Sessions | Tokens | Est. Avg Context/Session |
|---|
| Dec 14 | 27 | 1.08M | 2.7M |
| Dec 24 | 14 | 835K | 4.0M |
| Jan 11 | 22 | 1.38M | 4.2M |
| Jan 17 | 16 | 104K | 437K |
| Jan 19 | 83 | 499K | 398K |
| Jan 20 | 31 | 181K | 5.8K |
8. Implementation: Context Tracking
New Files Created
hooks/lib/context-tracker.ts
- Estimates tokens from transcript content (~0.25 tokens/char)
- Records OpenTelemetry metrics
- Maintains historical data in
~/.claude/context-history.json - Tracks daily averages for trend analysis
New Metrics (exported to SigNoz)
| Metric | Description |
|---|
session.context.size | Estimated tokens at session start |
session.context.utilization | Context window % used (of 200K) |
session.starts | Session start counter |
New Trace Attributes
| Attribute | Description |
|---|
context.estimated_tokens | Token estimate |
context.utilization_percent | % of 200K window |
context.transcript_size | Raw transcript bytes |
context.message_count | Conversation turns |
context.is_resume | Whether session was resumed |
📊 Context: 45K tokens (22.5%)
[████░░░░░░░░░░░░░░░░] 🟢
9. Key Insights
Efficiency Gains
- 3x more messages in January with fewer tokens = significant efficiency gain
- Tokens per message dropped 69%: 122→38 tokens
- Tokens per session dropped 64%: 40K→14K
- Cost per session dropped 64%: $22.17→$7.90
Usage Patterns
- Research days have high tokens, low commits
- Implementation days have high messages, low tokens
- More sessions = smaller contexts (deliberate management)
- Cache hit ratio of 14.9:1 indicates excellent context reuse
Cost Optimization
- Cache operations = 97% of cost
- Without caching, input would cost ~$19,600 (5.3x more expensive)
- Net cache benefit: ~$16,000 saved via cache reads
10. Recommendations
- Continue short session strategy — Week 3 showed 81% cost reduction per session
- Track context utilization — New hooks will provide visibility
- Monitor research spikes — 7x research:implementation token ratio is high
- Consider Sonnet for subagents — 11x increase shows good delegation
Report generated: January 20, 2026 Data period: December 9, 2025 - January 20, 2026 Total analysis cost: Included in Jan 20 session metrics
