Mistral AI
Mistral-Large-4
What Mistral-Large-4’s own documents say it is built from — every method with the sentence that describes it.
- Parameters
- ~1T / ~52B active
- Context
- 1M
- Attention
- sparse (indexer)MLA converted to DeepSeek Sparse Attention (indexer top-k 2048), NoPE every 4th layer, YaRN
- Experts
- 128 routed + 1 shared, top-4 (first layer dense)
- Input
- text, image
- Licence
- Mistral Ecosystem Open-Weight License (modified MIT)
- Released
- 2026-10-29
Against the consensus recipe
Its documents state 8 of the 18 methods the field agrees on.
Data curation
- Benchmark decontaminationnot in its documents
- Synthetic data generationnot in its documents
- Corpus deduplicationnot in its documents
Training objective
- Multi-Token Predictionnot in its documents
- Knowledge insulationnot in its documents
- Quantization-aware distillationnot in its documents
Model architecture
- Mixture of expertscore
- Multi-Token Predictionnot in its documents
- Grouped-query attentionnot in its documents
Optimization
- Muonused
- Expert parallelismused
- Quantization-Aware Trainingnot in its documents
Post-training
Inference & serving
- Thinking modedefault
- Multi-Token Predictionnot in its documents
- FP8used
Checked in its code
Its code shows 8 of the 19 architecture features checked — 4 of them are not stated in its documents.
Sources say what a lab chose to describe; the modeling code says what the checkpoint runs. Each feature below was put to a code verifier, and only the config values and source lines it cited that were found in the files are shown, linked to the line.
Read from Hugging Face Transformers (modeling_mistral_large4.py, configuration_mistral_large4.py) and config.json at revision af26e7869a, by anthropic/claude-sonnet-5 on 2026-11-07.
| Feature | Code | Its documents | Verdict |
|---|---|---|---|
| Grouped-query attention | not in its code | not stated | num_attention_heads equals num_key_value_heads (64 both) and attention uses MLA-style latent KV rather than GQA with fewer KV heads. |
| Multi-head latent attention | in its code | adopted | Keys/values are compressed into a shared kv_lora_rank latent via kv_a_proj_with_mqa/kv_a_layernorm and up-projected per head by kv_b_proj… show evidenceKeys/values are compressed into a shared kv_lora_rank latent via kv_a_proj_with_mqa/kv_a_layernorm and up-projected per head by kv_b_proj, with a decoupled RoPE key, DeepSeek-V2 style. text_config.kv_lora_rank = 512text_config.qk_rope_head_dim = 64 |
| Sliding-window attention | not in its code | not used | No sliding_window key or windowed masking mechanism appears in the config or code; all layers are 'indexed_attention' with full causal… show the whole verdictNo sliding_window key or windowed masking mechanism appears in the config or code; all layers are 'indexed_attention' with full causal masking. |
| Interleaved sliding-window and global attention | not in its code | not stated | layer_types lists only 'indexed_attention' for every layer, with no sliding-window variant interleaved. |
| Indexer-selected sparse attention (DeepSeek Sparse Attention) | in its code | adopted | A MistralLarge4TextIndexer scores past tokens and the attention forward selects the indexer's top-k (index_topk) token indices to restrict… show evidenceA MistralLarge4TextIndexer scores past tokens and the attention forward selects the indexer's top-k (index_topk) token indices to restrict attention, DeepSeek Sparse Attention style. text_config.index_topk = 2048text_config.index_n_heads = 32text_config.index_head_dim = 128 |
| Linear-attention or state-space layers alongside full attention | not in its code | not stated | There is no linear-attention or SSM mixer implementation; every layer is the same MLA/DSA attention module per layer_types. |
| Gated DeltaNet layers | not in its code | not stated | No gated delta-rule recurrent mixer or related config keys (linear_num_key_heads, linear_conv_kernel_dim) exist in this model's code or… show the whole verdictNo gated delta-rule recurrent mixer or related config keys (linear_num_key_heads, linear_conv_kernel_dim) exist in this model's code or config. |
| Mamba-2 layers | not in its code | not stated | No Mamba-2/SSD state-space block implementation or related config keys (mamba_num_heads, ssm_state_size) are present. |
| Learnable attention sink | not in its code | not stated | There is no learned per-head sink logit added to the attention softmax denominator anywhere in the attention or eager_attention_forward… show the whole verdictThere is no learned per-head sink logit added to the attention softmax denominator anywhere in the attention or eager_attention_forward code. |
| Gated attention output | not in its code | not stated | The attention output is reshaped and passed directly to o_proj with no sigmoid gating multiplication. show evidence |
| QK normalization | not in its code | not stated | Queries and keys in the main attention module are not normalized before the dot product; only the indexer's keys use a LayerNorm (k_norm)… show the whole verdictQueries and keys in the main attention module are not normalized before the dot product; only the indexer's keys use a LayerNorm (k_norm), which is part of the separate sparse indexer, not the main QK path. |
| Partial RoPE code only | in its code | not stated | RoPE is applied only to the qk_rope_head_dim slice (64) of each 128-dim qk head while qk_nope_head_dim (64) carries no rotary signal… show evidenceRoPE is applied only to the qk_rope_head_dim slice (64) of each 128-dim qk head while qk_nope_head_dim (64) carries no rotary signal, DeepSeek-style decoupled RoPE. text_config.qk_rope_head_dim = 64text_config.qk_nope_head_dim = 64 |
| YaRN RoPE scaling code only | in its code | discussed | rope_parameters.rope_type is 'yarn' and the rotary embedding module dispatches to the YaRN init function via ROPE_INIT_FUNCTIONS, with… show evidencerope_parameters.rope_type is 'yarn' and the rotary embedding module dispatches to the YaRN init function via ROPE_INIT_FUNCTIONS, with mscale applied to attention scaling. text_config.rope_parameters = {"attention_factor": 1.0, "beta_fast": 32, "beta_slow": 1, "factor": 2048, "o… |
| Layers without positional encoding (NoPE) code only | in its code | not stated | no_rope_layers marks every 4th softmax-attention layer (value 0) to skip RoPE entirely while other layers apply rotary embeddings, all… show evidenceno_rope_layers marks every 4th softmax-attention layer (value 0) to skip RoPE entirely while other layers apply rotary embeddings, all within the same indexed_attention layer type. text_config.no_rope_layers = [1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1… |
| Mixture of experts | in its code | adopted | MistralLarge4TextMoE routes tokens to n_routed_experts=128 experts via a top-k router selecting num_experts_per_tok=4 experts per token… show evidenceMistralLarge4TextMoE routes tokens to n_routed_experts=128 experts via a top-k router selecting num_experts_per_tok=4 experts per token, applied in all 'sparse' MLP layers. text_config.n_routed_experts = 128text_config.num_experts_per_tok = 4 |
| Shared expert | in its code | adopted | MistralLarge4TextMoE adds a shared_experts MLP sized by moe_intermediate_size * n_shared_experts=1 that every token passes through in… show evidenceMistralLarge4TextMoE adds a shared_experts MLP sized by moe_intermediate_size * n_shared_experts=1 that every token passes through in addition to routed experts. text_config.n_shared_experts = 1 |
| Auxiliary-loss-free load balancing (selection bias) | not in its code | not stated | The router (MistralLarge4TextTopkRouter) computes softmax scores and selects top-k directly with group masking but applies no learned… show evidenceThe router (MistralLarge4TextTopkRouter) computes softmax scores and selects top-k directly with group masking but applies no learned per-expert selection bias (no e_score_correction_bias in config or code). |
| Multi-token prediction layers | not in its code | not stated | The code explicitly skips MTP weights ('mtp_dspark' keys ignored on load) and the config has no num_nextn_predict_layers/mtp_num_layers… show evidenceThe code explicitly skips MTP weights ('mtp_dspark' keys ignored on load) and the config has no num_nextn_predict_layers/mtp_num_layers count declaring MTP modules to build. |
| SwiGLU feed-forward code only | in its code | not stated | Both the dense MLP and MoE experts compute silu(gate_proj(x)) * up_proj(x), i.e. SwiGLU, with hidden_act set to silu. show evidencetext_config.hidden_act = "silu" |
token mixer 10
softmax attention 2
“LeRoPE and SSMax let the model learn rotary frequencies and per-head query scaling.”
multi-head latent attention 1
“Its attention blocks combine Multi-head Latent Attention (MLA) with a DeepSeek Sparse Attention (DSA) indexer”
sparse attention 8
“Mistral Large 4 combines a large pool of experts with sparse attention and a Pixtral vision encoder.”
“FlashMLA sparse attention and an FP8 latent KV cache (fp8_ds_mla) by default”
“we adapted ML4 to DSA, using lightweight indexers to select the top-2,048 key tokens for each query”
sparse attention indexer 5
“MLA keeps a compressed KV representation, and DSA selects up to 2,048 key positions for each query.”
“MLA attention with a DSA sparse indexer (top-2048 keys)”
“we warm up the indexers for approximately 15.1B tokens through KL-based distillation of the main model’s attention distributions”
“We therefore share a single DSA indexer among the former SWA layers within each of these groups.”
“using lightweight indexers to select the top-2,048 key tokens for each query”
channel mixer 7
mixture of experts 7
“Open-weight hybrid instruct-and-reasoning MoE with multimodal input”
“Open-weight hybrid instruct-and-reasoning MoE with multimodal input”
expert routing 1
“a softmax router selects four of the 128 experts for each token.”
expert load balancing 1
“we compute the load-balancing loss over each expert-parallel (EP) group”
fine-grained experts 2
“granular Mixture-of-Experts with 128 routed experts (4 active per token) and 1 shared expert”
“uses a granularity of 4 for its mixture-of-experts layers”
positional encoding 2
“Mistral uses LeRope, a YaRN-scaled RoPE with one learned multiplier per frequency.”
“For each decoder block, there’s a 3:1 RoPE to NoPE pattern, i.e. 3 decoder layers employ RoPE and every fourth layer uses NoPE.”
multimodal architecture 5
“our flagship 1-trillion-parameter, natively multimodal model”
“with multimodal input”
“Mistral Large 4 employs a 1.6B Pixtral encoder.”
“Pixtral-style vision encoder trained to handle variable image sizes and aspect ratios”
“It uses 14 × 14 patches and a patch-merge projector with spatial merge size 2.”
2 more mentioned or evaluated
positional encoding “we removed the attention windows and used YaRN to rescale the RoPE frequencies by 8×”
expert load balancing “Learn more about our load balancing and provider routing options.”
Not used 3
token mixer “Every other full-attention layer reuses the KV cache of the preceding full-attention layer.”
sliding window attention “the remaining layers use sliding-window attention (SWA) with a window size of 512 tokens”
multi-head latent attention “One in every four layers uses full attention, the remaining layers use sliding-window attention (SWA) with a window size of 512 tokens.”
distillation objective 1
“we optimize the following per-token objective over the full vocabulary”
auxiliary loss 1
“For a subset of data sources prone to over-optimization, we add a KL penalty to the reference policy using the k 2 estimator [40] with…”
optimization 19
optimizer 3
“We use Muon [38, 39] (Nesterov momentum 0.95, quintic Newton–Schulz with 10 iterations, spectral scaling) at a constant learning rate of…”
“We use the AdamW optimizer with (β1,β 2) = (0.975,0.975) during pretraining.”
“(Nesterov momentum 0.95, quintic Newton–Schulz with 10 iterations, spectral scaling)”
learning-rate schedule 1
“we ramp up the batch size in stages with a share of tokens that grows exponentially per stage”
training precision 3
“and blockwise scales in both the generator and trainer.”
“we use FP8 tensor cores for fine-grained low-precision training”
“A quantized NVFP4 checkpoint is also available, with a smaller memory footprint, faster inference and same performance.”
training stability 1
“router replay re-executes the expert-routing decisions made at inference time during the training forward pass, eliminating divergence…”
training parallelism 6
“with expert parallelism, +14–17% prefill req/s at 32k”
“To mitigate this, we make use of context parallelism (CP) [80] to shard and de-replicate the KV cache across the sequence dimension instead.”
“maximized compute–communication overlapping”
“By rewriting the CP collectives with latency-optimized exchange patterns, in-kernel overlap of compute and communication, and other fusion…”
“We accelerate training by leveraging ZeRO-2 optimizations”
“Prefill context parallelism splits each prompt across the 8 ranks”
training runtime 5
“the NVLink-C2C link to the Grace CPU for fast activation CPU offloading”
“asynchronous workloads to minimize impact on actual training (data loading, checkpointing, metrics, etc.)”
“automated health checks and crash analysis help identify failures and coordinate interventions across time zones.”
“fine-grained activation recomputation”
“We implement a custom fused kernel for fast full-logit KL loss backward”
data filtering 1
“we perform a comprehensive set of checks, including hackability, problem–verifier mismatch, and insufficient learning signal”
synthetic data 7
“we invested heavily in a high-quality, scalable agent-based pipeline to generate synthetic data”
“We provide the authoring agent with a minimal set of tools to perform all necessary procedures to create realistic, high-quality tasks”
“A refinement agent runs several solvers of varying capabilities against each task and inspects their trajectories”
“We also include the successful authoring agent trajectories from these pipelines in our midtraining corpus”
“we collect millions of technical seeds from our pretraining corpus”
“By adopting taxonomies, we have fine-grained control over the coverage and complexity of the generated synthetic data.”
“We use a two-stage approach: first, an authoring agent turns each candidate into a runnable environment”
data mixture & curriculum 2
“We optimize a highly heterogeneous training mixture covering multimodal, alignment, instruction-following, cybersecurity, code agents…”
“The pipeline starts with curriculum-driven data loading”
sequence packing 1
“packing=True,”
tokenization 1
“applies trie-based tokenization to handle possible re-tokenization events”
1 more mentioned or evaluated
data sourcing “We plan to move towards an internal, offline version of the web to improve control over results and mitigate safety risks.”
supervised fine-tuning 6
“the final model owes them to both post-training stages (supervised fine-tuning and RL)”
“peft_config=LoraConfig(”
“assistant_only_loss=True,”
“frozen throughout pretraining and unfrozen only for supervised fine-tuning (SFT)”
“The model is free to use, modify, fine-tune, and commercialize”
“with TRL you could fine-tune smaller models from Mistral Large 4 traces patching your code base.”
reinforcement learning algorithm 14
“We use a CISPO-style [34] REINFORCE [35] loss.”
“The objective corrects for this policy lag and numerical mismatch between the inference engine and the trainer by importance weighting…”
“Unlike PPO [36], the ratio is not clipped inside the objective, so tokens still contribute to learning when their weights are clipped.”
“Within a group, every trainable token gets the same weight, scaled so that the group’s total weight equals its number of rollouts”
“clipped from above at 3 (with no lower bound) and treated as a constant via a stop-gradient.”
“we've put that feedback to work while continuing to scale reinforcement learning, driving massive improvements in coding and defensive…”
“When every rollout in a group gets the same reward, all its advantages are zero and it carries no learning signal. We drop such groups…”
“A rollout’s advantage is its reward minus the group mean, and advantages are standardized across the batch.”
“For each prompt, we sample a group of rollouts. Group size is set per data source: 16 by default, 32 for agentic coding, and between 2 and…”
“We first subtract the groupwise mean reward, then normalize the resulting centered rewards by their standard deviation across the entire…”
“For multi-turn trajectories, we additionally apply MiMo-style segment-level advantage correction [37] to turns flagged for generation…”
“Novel methods and optimizations across both stages minimize off-policy drift and enable stable RL over long horizons.”
“Our RL pipeline enabled us to reach 84.5% on BrowseComp”
“starting from an SFT checkpoint that was achieving 75%”
reward modelling 12
“Each data source defines its own reward function by combining signals into a composite reward.”
“Ambiguous cases default to the agent’s fault, so unreliable behavior is never silently dropped and implicitly rewarded.”
“We also employ an adversarial agent to check for reward-hacking vulnerabilities”
“with reward models, unit tests, LLM judges, and static checks combined as needed for each task”
“A candidate patch receives positive reward only if it completely neutralizes the exploit while preserving existing test-suite passes.”
“Tests that fail before the reference solution and pass afterwards become the verifier”
“we augment the task reward with a group-relative length bonus for correct rollouts.”
“which applies pairwise rewards and groupwise penalties”
“we quantize rollout lengths into bins of width ∆ = 0.1Lmin, anchored at Lmin.”
“we design task rewards to have broadly comparable scales, with values typically in [0,1].”
“checks its answers against a standard chemistry tool.”
“reproduce-then-patch tasks use a verifiable reward, so it fits with RL environments that could be used to further RL smaller security…”
policy distillation 4
“We combine these teachers into a single model through multi-teacher on-policy distillation [45, 46].”
“Thanks to multiple-teacher distillation from checkpoints with different levels of verbosity”
“We implement a remote teacher infrastructure so teachers can scale across separate GPU nodes, decoupling the number of teachers from…”
“For the final release model, we initialize the student from the high-reasoning RL checkpoint and train it with three teachers, yielding a…”
rollout & RL infrastructure 19
“We use a custom, fully asynchronous RL training pipeline implemented in Ray”
“All of our environments, from single-turn question answering (QA) to long-horizon agentic coding, are defined against a single interface.”
“The episode generation concurrency adapts following backpressure signals from generators, services, and the trainer, which limits staleness.”
“Our pipeline feeds RL instances through a series of actor pools that handle agentic-loop rollout, verification, scoring, and tokenization.”
“an autoscaling fleet of actors generates tens of thousands of rollouts in parallel while model training proceeds asynchronously”
“All supporting services autoscale to adapt to the load imposed by in-progress rollouts.”
“admission of groups in the runner pool depending on maximum concurrency and backpressure signals”
“A centralized bin-packer then packs individual trajectories into fixed-length microbatches sent to each data-parallel rank.”
“rollout failures due to a sandbox crash or service call error are discarded in RL and retried in evaluation.”
“Auxiliary model calls made inside the environment, such as user models and feedback digesters, are marked to avoid training on them. Only…”
“The trainer takes an optimizer step once it has accumulated a fixed number of usable trajectories (7k for ML4)”
“supporting rollout budgets of millions of tokens across multiple compactions while keeping staleness low”
“In-flight rollouts pause, load the new weights, and resume without recomputing their KV cache.”
“Generators are served on Ray through vLLM engines [33] and receive weight updates via NCCL at each training step.”
“with higher priority for sibling rollouts of started groups”
“we can recover by routing to another healthy provider”
“We reduce the communication cost between teachers and trainers by transferring only hidden states via NVIDIA Inference Xfer Library (NIXL)…”
“A shared, composable interface allows a single training run to combine tasks ranging from single-turn chat and complex scientific problem…”
“In environments with a submit tool, the per-turn verification is skipped and only submission ends the episode.”
agentic post-training 2
“Agentic tasks share a common loop: the agent acts until it produces a final message without tool calls. A verifier then scores the state…”
“use solver pass rates to calibrate the difficulty of each instance”
decoding strategy 8
“Both checkpoints ship the DSpark drafter. The speculative depth is fixed at 3 by the checkpoint.”
“propose a block of 3 tokens”
“with a recurrent Markov head”
“3 dense MLA draft layers in BF16 that propose a block of 3 tokens”
“for all reasoning efforts, we recommend the following sampling parameters: temperature=1.0 top_p=0.95”
“Together with block verification [84], DSpark improves latency and throughput by 90% compared to a baseline without speculative decoding”
“we use a recurrent conditioning head and unique mask tokens [83], which improve acceptance length by +1.2% and +0.5%, respectively”
“we use a recurrent conditioning head and unique mask tokens [83], which improve acceptance length by +1.2% and +0.5%, respectively”
reasoning control 2
“three reasoning efforts — low ,medium ,high (default)”
“reasoning cannot be disabled, so using reasoning_effort: none would still use high.”
KV cache management 4
“cross-layer KV sharing”
“Prefix caching (on by default) for agentic workloads”
“Full-attention layers in the second half of the transformer attend to the KV cache produced by the first half instead of maintaining their…”
“Decode context parallelism shards the latent KV cache across the 8 ranks”
inference quantization 8
“the latent KV cache is plain FP8 (E4M3) by default, 1.86x the capacity of bf16”
“we adopt FP8 inference to improve generation speed and match the trainer”
“We also quantize ML4’s routed experts into NVFP4 format while keeping other components at their training precision, resulting in a mixed…”
“We found that naive abs-max quantization already suffices and matches the FP8 reference in all evaluations with appropriate calibration.”
“In combination with a native FP8 KV cache and offloading, this gives us approximately 20 million tokens of in-memory KV-cache space on…”
“The routed experts run as OCP MXFP4 on AITER's fused MoE”
“We use 200M tokens from our SFT mixture for calibration.”
“from the FP8 checkpoint re-quantized at load (VLLM_FP8_MOE_REQUANT=mxfp4, layer by layer, so peak load memory stays low)”
serving parallelism 2
“Disaggregated 4+4 beats colocated 8-GPU serving by 1.8–2.2x at 8k and 4–5x at 32k/60k”
“The recipe runs vLLM with the FP8 weights split across all eight GPUs using tensor and expert parallelism”
inference scheduling 2
“We autoscale our inference fleet based on live request and server metrics.”
“Route each session to the same DP rank on DP sides”
inference kernel 5
“vLLM serves the model on model runner V2 (the default) with breakable CUDA graphs”
“systematic kernel fusing”
“We handle this setting correctly while maintaining efficiency by leveraging blockwise attention from FA4 [41].”
“Runs every MoE layer of a 1-, 2- or 4-request decode step as one persistent FlyDSL kernel”
“attention, indexer, router, MXFP4 experts and both all-reduces in one launch”
agentic scaffolding 5
“Mistral Large 4 enforces response formats natively, including strict JSON schemas.”
“We are now rebuilding our sandbox infrastructure around microVMs to strengthen isolation and support efficient snapshotting.”
“keep the set of tools well-defined and limit their number to the minimum required for the use case”
“define a clear environment and use case, including guidance on how to effectively leverage tools in agentic systems”
“Environments use this hook for multi-turn interaction: for example, feeding back a digest of failing tests as a hint, moving the agent on…”
1 more mentioned or evaluated
inference scheduling “Learn more about our load balancing and provider routing options.”
inference engine 2
“We handle inference at scale by building on llm-d [86, 87] for routing and Model Express [88] for faster startup.”
“the open source TensorRT LLM library for high-throughput serving.”
training framework 1
“We also ship day 0 support of fine tuning Mistral Large 4 using TRL.”
kernel & quantization library 3
“the DSA indexer on AITER's gfx950 indexer kernels”
“The sparse MLA runs on AITER's MLA decode kernels (ROCM_AITER_MLA_SPARSE)”
“The preset adds EP, as used by the PCP prefill side of the P/D presets”
infrastructure service 4
“Our architecture is multi-cluster by design, and leverages a central scheduler with the option to spill over to cloud providers when…”
“we use Model Express for P2P checkpoint transfer over InfiniBand and warm up cache sharing.”
“Routed with the NIXL toy proxy (tests/v1/kv_connector/nixl_integration/toy_proxy_server.py), not vllm-router.”
“We use the Ray data plane to transport heavy objects across the pipeline, such as MoE routing decisions for router replay.”
cybersecurity - vulnerability discovery; exploit development 2
“Our model’s built-in guardrails distinguishes between defensive and malicious, offensive cybersecurity tasks.”
“We automatically construct rigorous test suites that probe the affected paths using both adversarial exploits and standard inputs”
evaluation harness 1
“A test harness is validated only when it triggers on the unpatched codebase and passes following the reference fix.”
human & real-world evaluation 1
“uses it in a browser to catch its own bugs and slowdowns, then fixes what an independent reviewer flags until all requirements pass.”
other 3
filed at the root 3
“Every request or response is logged to an append-only list of records, which is the single representation of an episode, from which…”
“we added KV-producer weights by copying them from the last long layer that maintained its own KV weights”
“The goal was to remove KV-cache compression mechanisms to achieve an architecture as close as possible to a full-attention model.”
unfiled 14
“customers are tailoring Mistral Large 4 with proprietary security data, policies, detection logic, and threat environments using Mistral…”
“The records also support replay, which lets us check that refactors do not change environment behavior.”
“Improvements primarily come from hardware multicast, latency-first data-exchange patterns, removing in-kernel synchronization, and fusing…”
“We actively monitor our RL training runs for reward hacking and environment leaks.”
“a multi-agent setup inspects each workspace for realism, consistency, and file integrity, before providing integrated feedback”
“We monitor the quality of the learning signal and efficiency of the online training pipeline.”
“we overlap the teacher forward pass with the final batch of the trainer forward pass.”
“Organizations can customize Mistral Large 4 around their proprietary data with Mistral Forge”
“preview access to a specialized variant of Mistral Large 4 optimized for advanced cybersecurity workflows”
“Mistral’s Applied AI experts can work alongside customer teams to build and deploy these tailored systems.”
“For organizations that prefer a managed service, Mistral Large 4 is available through Microsoft Foundry, with private deployment options…”
“Developers can use the open source NeMo AutoModel library for customization”
“Organizations have the ability to inspect, customize, and deploy the model on their own infrastructure.”
“For well-defined cybersecurity tasks, it can be further optimized into smaller, task-specific variants for greater accuracy”
Sources
The curated source list for this model. "not fetched" means the URL is recorded but its text was not read in the current run.
| Source | Kind | Publisher | Status |
|---|---|---|---|
| mistral.ai/news/mistral-large-4-oss | official blog | Mistral AI | read |
| mistral.ai/news/mistral-large-4 | official blog | Mistral AI | read |
| huggingface.co/mistralai/Mistral-Large-4-1T-A52B/blob/main/assets/techn… | technical report | Mistral AI | read |
| huggingface.co/mistralai/Mistral-Large-4-1T-A52B | model card | Mistral AI | read |
| huggingface.co/blog/mistral-large4 | third party analysis | Hugging Face | read |
| recipes.vllm.ai/mistralai/Mistral-Large-4-1T-A52B | vendor docs | vLLM | read |
| docs.mistral.ai/models/mistral-large-4-0 | vendor docs | Mistral AI | not fetched |
| openrouter.ai/mistralai/mistral-large-4-0 | vendor docs | OpenRouter | read |