A bigger context window alone
Relying on a 1M-token model and doing nothing else. Keep everything in the window and hope the model finds it.
Verdict
Buys time, not memory; research on long histories shows models with 1M-token windows struggle as conversations lengthen.
Best for
Short-to-medium sessions where the whole task fits comfortably in the window.
Poor fit
Anything that outlives one session, or a history bigger than the window.
Scores
| Criterion | Score | Why |
|---|---|---|
| Survival across sessions | Weak | Ends with the session. |
| Cross-tool reach | Weak | One model, one session. |
| Multi-agent sharing | Weak | Not shared. |
| Retention of reasons | Partial | Kept verbatim, but only while they fit and only if the model attends to them. |
| Scale | Weak | Fails beyond the window; degrades before it (BEAM paper). |
| Openness | Weak | Depends entirely on the model provider. |
Strengths
- No setup and no loss while everything fits.
- The model sees the original words, not a summary.
Limits
- Ends with the session; nothing carries to the next one.
- Once the window fills, it falls back to compaction.
- Retrieval inside a very long context is unreliable; the BEAM paper reports models with 1M-token windows struggling as dialogues lengthen.
How it fails
- The answer is in the window but the model misses it among a million tokens of tool output.
- Every call re-sends the whole history, so cost grows with the run.
Cost
Priced per token by the model provider; long contexts cost more per call. Check your provider's pricing page.