Frontier AI, out-remembered.

The benchmark is MRCR v2 8-needle: a long conversation full of requests that look alike, where the model has to resolve a reference back to the right earlier one. Native Context Memory is model-agnostic.

1M-token context

  1. 1DeepSeek-V4-Flash + Native Context Memory99.9%Adaptiq
  2. 2Meta Muse Spark 1.3 · Max98.1%Meta
  3. 3OpenAI GPT-6 Astra96.3%OpenAI
  4. 4DeepSeek-V4-Flash · Max78.7%DeepSeek
  5. 5Claude Opus 4.678.3%Anthropic
  6. 6Gemini 3.7 Flash62.5%AI Release Tracker
  7. 7Qwen3.8-Flash-Next26.4%Qwen

5M-token context

  1. 1DeepSeek-V4-Flash + Native Context Memory98.5%Adaptiq

5M is just the start.

A 5M-token context in each agent session is the foundation we build on. We believe unlimited context is achievable, and that is what we are working towards.

What Native Context Memory adds.

We ran MiMo V2.6 Flash through all ten MemoryAgentBench tasks, with and without NCM. MemoryAgentBench measures whether an LLM can remember over time: not whether it can read a long text once, but whether it can store, update and retrieve what it has been told in pieces along the way.

It keeps track of what changed.

When a fact is replaced by a newer one, MiMo on its own often misses the change. With NCM it goes with the newest.

It connects what was said at different times.

How many, how long ago, which came first: when the answer is spread over several conversations, a model alone loses the thread. NCM finds each piece and puts them together.

It stays exact, however much there is to remember.

On its own, MiMo’s answers vary from one long text to the next. With NCM they hold steady, because the answer comes from the lines themselves.

MemoryAgentBench

Conflict resolution

+68.6Conflicting facts, multi-hopFC-MH
+16.0Conflicting facts, single-hopFC-SH

Accurate retrieval

+23.9Multi-hop retrievalMH
+10.7Long-term chat memoryLongMemEval
+9.2Event recall in novelsEventQA
+9.0Single-hop retrievalSH

Test-time learning

+9.9In-context classificationMCC
+1.1Movie recommendationsMovieRec

Long-range understanding

+7.0Detective novel reasoningDetQA
−6.5Book summarizationInfSum
MiMoMiMo + NCM

Tokens used

MiMo alone used 3.5 times more new input tokens than MiMo with NCM: 179.0 million against 51.5 million across all 2,071 questions.

MiMo aloneMiMo + NCM
New input tokens179.0M51.5M
Input read from cache639.7M600.0M
Total input818.7M651.5M
Output tokens0.60M5.25M
New input tokens per questionabout 86,400about 24,900
Total cost (OpenRouter)$27.01$10.36

What would you do with more context?

The benchmark shows how much a model can hold on to. What that is worth depends on what you give it, so try it on your own work.