The 2.80 GB Q4_K_M file fits by size inside each pool, but that is not a compatibility or speed result. Microsoft evaluated the source checkpoint with SGLang on GPU servers with the Leaf harness, not this third-party quant. The official card says its BF16 weights alone are about 9.32 GB and that memory rises substantially near the evaluated 131K-token context.
Exact Q4_K_M file at revision dfbb801. Remaining-space figures are arithmetic, not measured runtime requirements.
Memory pool
Exact weight file
Space after weights
What is verified
4 GiB
2.80 GB / 2.61 GiB
1.39 GiB
The file fits. Runtime, context cache, operating system and tools still need memory; usable full offload was not measured.
8 GiB
2.80 GB / 2.61 GiB
5.39 GiB
More arithmetic headroom, but no FrogNano token-speed or long-context result was verified for an 8 GiB device.
16 GiB
2.80 GB / 2.61 GiB
13.39 GiB
More room for runtime and context than the smaller pools; this still does not verify backend support, maximum context or speed.
Formula: pool GiB − 2,797,526,816 file bytes ÷ 1,073,741,824. System RAM and dedicated VRAM are not interchangeable, and shared-GPU allocation is not inferred from installed RAM.
Start with a short context and the controlled Leaf-style tool loop. The official evaluation's roughly 131K-token context can require substantially more memory than the weight file.
Measure peak memory, prompt processing and generation speed on your own workload before treating any computer as suitable.
Microsoft did not validate quantized formats, CPU-only servers, desktop operating systems or mobile devices for this release. The official card also says FrogNano is text-only for its intended and evaluated use, even though inherited visual components exist. Generated patches require human review and security testing.
Exact computers to compare
These are purchase configurations from the hardware catalog, not FrogNano test results. The 16 GB Windows mini PC would be a CPU-inference planning comparison unless its integrated-GPU allocation and backend are separately verified; the 24 GB Mac uses unified memory but this exact model and quant were not tested here.
These scores were compiled from the linked sources. These published scores were not independently tested here. Different prompts, precision, thinking budgets and tools can change results.
SWE-bench Verified · avg@3
61.5 / 100
Software issue resolution averaged over three attempts. Agent setup affects scores; kept separate from other SWE-bench results.
Benchmark efficiency: 21.984 published score points per GB of selected weights. This ratio is not an inference-speed measure; check the test settings to see whether the score covers the selected quantization.
Test settings: Microsoft publisher evaluation of the source checkpoint on all 500 SWE-bench Verified issues using the Leaf harness, 150 interaction steps, about 131K combined context, temperature 0.6 and repetition penalty 1.0; resolution rate averaged across three seeds.
Publisher-run source-checkpoint result, not an independent test of the selected Q4_K_M file. Microsoft states that quantized formats, CPU-only servers, desktop operating systems and mobile devices were not validated for this release. Agent scaffold, context, tools and runtime can materially change results.
Context and version evidence: The repository and file name identify the selected model. Revision hashes are shown above when recorded. Tested context length, full hardware configuration and run date are not consistently recorded in this snapshot. Only settings explicitly stated in each benchmark record are known.
The 2.798 GB figure covers model weights in this file. It excludes the context cache, runtime, operating system and any separate multimodal files. It is not a measured RAM requirement or a token-generation speed.
These complete configurations pass the initial weight-size check for the selected 2.80 GB file. Allow extra memory for context and software. Check the model’s runtime notes and your inference software before buying.
Listing evidence: 2026-09-25. Amazon.com base price visible in an owner-supplied screenshot reviewed 25 Sep 2026; capture time not supplied. Dated reference, not a live quote. Excludes card signup offers, coupons, tax and shipping.
Listing evidence: 2026-10-03. Apple US launch price announced 22 September 2026: USD 1,699. Dated manufacturer reference, not an Amazon offer; tax and shipping may differ.
Listing evidence: 2026-09-24. GMKtec, EVO-X2 / 64GB / 1TB, recorded 22 Sep 2026; exact ASIN capacity identified 24 Sep. This is a dated reference, not a current Amazon price; tax, shipping and bundle contents may differ.
Memory matches are estimates, not tested compatibility or speed. As an Amazon Associate I earn from qualifying purchases.
Confirm model architecture and quantization support in your runner. The selected weights use 2.798 GB; this excludes context cache, runtime buffers and other programs.
These are the first listed tiers that pass our reserve policy, not measured minimum requirements. System RAM and VRAM remain separate. Use the memory tool to change the reserve for your context and runner.