{"slug": "mistral-nemo-12b-cheapest-dekallm-0-02-m-input", "title": "Mistral Nemo 12B - cheapest: DekaLLM $0.02/M input", "summary": "DekaLLM offers the cheapest inference for Mistral Nemo 12B at $0.02 per million input tokens, delivering 3,444 points per dollar based on a general score of 62. The 12.2-billion-parameter model supports a 128k context window and is released under Apache 2.0 license as of July 2024.", "body_md": "# Mistral Nemo 12B\n\nconsumer- 12.2B\n- 128k\n- apache 2.0\n- Jul 2024\n\n## Scores\n\n## Score per dollar\n\n3444 pts per $/M input\n\ngeneral_score (62) divided by cheapest input price\n($0.02/M).\nHigher is better value. [See live pricing](/models/mistral-nemo-12b/pricing).\n\n## Run it locally\n\nPer-quant memory needs and a static \"can you run it?\" reference - no rig entry required\n\n### Can you run it? - reference rigs\n\n| Rig | Q4_K_M | Q8_0 |\n|---|---|---|\n| 4x H100 80GB (320GB) | fast 1009.4t/s | fast 558.3t/s |\n| 4x RTX 5090 (128GB) | fast 539.9t/s | fast 298.6t/s |\n| 4x RTX 4090 (96GB) | fast 303.7t/s | fast 168.0t/s |\n| 2x RTX 5090 (64GB) | fast 270.0t/s | fast 149.3t/s |\n| 2x RTX 3090 (48GB) | fast 141.0t/s | fast 78.0t/s |\n| Single RTX 5090 (32GB) | fast 135.0t/s | fast 74.7t/s |\n| Mac Studio M4 Ultra 192GB | fast 89.7t/s | fast 49.6t/s |\n| Mac Studio M4 Ultra 512GB | fast 89.7t/s | fast 49.6t/s |\n| Single RTX 4090 (24GB) | fast 75.9t/s | fast 42.0t/s |\n| MacBook Pro M5 Max 128GB | fast 50.5t/s | fast 27.9t/s |\n| Single GTX 1080 Ti (11GB) | fast 36.5t/s | offload |\n| DGX Spark 128GB unified | fast 20.6t/s | ok 11.4t/s |\n| Epyc + 512GB DDR4-3200 + 2x RTX 3090 | ok 15.4t/s | ok 8.5t/s |\n| Epyc + 512GB DDR4-2400 + 2x RTX 3090 | ok 11.6t/s | slow 6.4t/s |\n| NVIDIA Jetson Orin NX 16GB | slow 7.7t/s | slow 4.3t/s |\n\nFit tiers use the same will-it-run logic as the rig finder. For comfortable fits, the badge reflects decode speed: fast >=20 t/s, ok 8-20 t/s, slow <8 t/s. t/s is a bandwidth estimate, not a measured benchmark.\n\n**How can a 24GB GPU run a 744B model?** It does not load the model into VRAM. The quantized weights (e.g. ~410GB at Q4) sit in system RAM; the GPU only holds the small shared attention and router tensors and accelerates prompt processing. Because GLM 5.2 is a Mixture-of-Experts model, each token activates only ~40B of its 744B params, so llama.cpp streams just those active experts from system RAM to the GPU each token (the `-cmoe`\n\noffload path).\n\nThat makes **decode speed bound by system-RAM bandwidth, not GPU bandwidth** - single digits on DDR4, which is why these rigs show 3-8 t/s even though they “fit.” A bigger GPU (e.g. 2x 3090) keeps more experts resident on-card and raises tok/s; a smaller GPU still runs it but pays the bandwidth tax. A 744B dense model could not run this way - only MoE’s small-active-params trick makes it possible.\n\nAggressive quants (1-2 bit) trade accuracy for size - roughly 17% accuracy loss at 2-bit vs full precision, and real long-context work often needs Q5 or Q6 even when lower quants “fit.”\n\nFormula estimates here are conservative; real tuned setups can exceed them (one HN user reports ~6 tok/s on a 512GB DDR4 + 2x 3090 rig).\n\n## Download options\n\n## Or run it in the cloud\n\nLive per-provider pricing, throughput and uptime - refreshed about 1 hour ago via OpenRouter. Click a column to sort.\n\n| Provider | Type | Input $/M | Output $/M | Cache $/M | Tok/s | Latency | Uptime | Value |\n|---|---|---|---|---|---|---|---|---|\n|\nDeepInfra\n|\nAPI | 0.02 | 0.03 | - | - | - | 99.95% | best uptime |\n|\nMistral\n|\nAPI | 0.15 | 0.15 | 0.015 | - | - | 99.94% | |\n|\nDekaLLM\n|\nAPI | 0.02 | 0.03 | - | - | - | 98.03% | cheapest |\n|\nNovita\navoid\n|\nAPI | 0.04 | 0.17 | - | - | - | 69.74% |\n\nDefault order: throughput among 95%+ uptime providers, then latency; subscriptions last. Sort by any column. Subscription rows show $/mo in the Value column - per-token columns are \"-\". Affiliate links are marked sponsored / nofollow. Confirm current pricing on the provider's site before committing.\n\n[Detailed API pricing page + JSON endpoint →](/models/mistral-nemo-12b/pricing)\n\n[See who runs Mistral AI in production →](/adoption/mistral)\n\n## Inference cost over time\n\nData accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.", "url": "https://wpnews.pro/news/mistral-nemo-12b-cheapest-dekallm-0-02-m-input", "canonical_source": "https://tokenstead.ai/models/mistral-nemo-12b", "published_at": "2026-07-21 21:08:41+00:00", "updated_at": "2026-07-21 21:25:42.000444+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-infrastructure", "ai-tools"], "entities": ["Mistral Nemo 12B", "DekaLLM", "DeepInfra", "Mistral", "Novita", "OpenRouter"], "alternates": {"html": "https://wpnews.pro/news/mistral-nemo-12b-cheapest-dekallm-0-02-m-input", "markdown": "https://wpnews.pro/news/mistral-nemo-12b-cheapest-dekallm-0-02-m-input.md", "text": "https://wpnews.pro/news/mistral-nemo-12b-cheapest-dekallm-0-02-m-input.txt", "jsonld": "https://wpnews.pro/news/mistral-nemo-12b-cheapest-dekallm-0-02-m-input.jsonld"}}