Pin official source and exact weights
Mutable model names are not immutable checkpoints.
Record exact 40-character SHA, file inventory, config and runtime.
Unpinned dependencies can change outputs.
CONTINUITY RESEARCH · FIELD VERIFIED
Primary sources last checked on 17 Nov 2026. Verification freshness is tracked separately from factual change, and unknown values remain explicit rather than inferred.
MODEL INTELLIGENCE · SOURCE-FIRST
Open Model Weights records Tess-4.5-594B-A17B as a 594B logical / 17B active (publisher); 383.07B packed stored tensor count-parameter agentic coding fine-tune open-weight model in the Agentic coding fine-tune family, with Continuity Research listed as the developer. The checked source repository provides 26 recognized weight artifacts, format evidence for Safetensors and a verified context window of 1,048,576 tokens. Values that are not evidenced in the checked sources remain explicitly undisclosed or unknown.
The repository declares apache-2.0. Commercial use allowed under apache-2.0 conditions. Open Model Weights reports the checked license evidence and classification; this is not legal advice.
The checked sources declare thinkingmachines/Inkling-2-Small as lineage evidence. Training recipe status: training-described-in-model-card; data disclosure status: not-disclosed-in-standard-metadata-or-obvious-model-card-section.
+ Save to my shortlistMonitor changes →Local browser monitoring · no background email alerts
DIRECT ANSWER · FEATURED MODEL BRIEFING
Source-grounded editorial layer · reviewed 2026-11-17
Tess-4.5-594B-A17B is Migel Tissera's agentic coding post-training derivative of Thinking Machines Lab's Inkling 2 Small. The publisher describes 594B logical total and 17B active parameters, while the inspected packed checkpoint's Safetensors metadata totals about 383.07B stored tensor values; those are different accounting measures. The official repository exposes 24 merged weight shards, an index and a speculative-decoding artifact, and the author separately distributes an approximately 327 MB LoRA adapter. The pinned card declares Apache-2.0 without a separately verified standard license file. Publisher results are not an independent OMW benchmark.
WHY IT MATTERS
Tess 4.5 targets agentic software engineering and tool-mediated coding, built by post-training the already public Inkling 2 Small architecture. The release is not a newly pretrained 594B base model and should remain a Notable derivative rather than an independently established overall frontier leader.
The model inherits the 594B-total, 17B-active logical architecture from Inkling 2 Small. The inspected Safetensors metadata counts about 383.07 billion stored values due to mixed packing and quantization; equating that raw total with the logical model size would misrepresent the architecture.
The pinned official repository exposes 24 Safetensors checkpoint shards, a weight index and mtp.safetensors for the merged derivative. The separately linked Tess LoRA adapter is about 327 MB, whereas the published merged checkpoint is described as approximately 394 GB.
MODEL-SELECTION VIEW
Tess 4.5 targets agentic software engineering and tool-mediated coding, built by post-training the already public Inkling 2 Small architecture. The release is not a newly pretrained 594B base model and should remain a Notable derivative rather than an independently established overall frontier leader.
According to the author, training used about 46.9 million tokens from engineering sessions and updates concentrated on attention projections and the output head. Routed experts remained frozen to avoid requantizing the packed expert weights.
The model inherits the 594B-total, 17B-active logical architecture from Inkling 2 Small. The inspected Safetensors metadata counts about 383.07 billion stored values due to mixed packing and quantization; equating that raw total with the logical model size would misrepresent the architecture.
The source config reports a 1,048,576-token window. That configured maximum is not a claim that every runtime, prompt shape or GPU configuration can execute one million tokens at acceptable latency.
The pinned official repository exposes 24 Safetensors checkpoint shards, a weight index and mtp.safetensors for the merged derivative. The separately linked Tess LoRA adapter is about 327 MB, whereas the published merged checkpoint is described as approximately 394 GB.
Adapters and full merged weights are different download paths. The author recommends around 500 GB or more aggregate accelerator memory for serving; this is a publisher deployment recommendation, not an OMW hardware measurement.
The developer reports 88.2% versus 86.3% for the underlying Inkling model on a 51-task SWE-Bench Pro v2 Hard subset and approximately nine percent fewer output tokens. The narrow study is interesting but should not be presented as a universal or independent win.
No independent Open Model Weights comparison of the exact Tess merged revision, the LoRA and the original Inkling baseline has been run. Reproduction requires identical prompts, effort settings, coding tools, dataset, judge and runtime.
The publisher declares Apache-2.0 on the model card and repository metadata. No separate standard license file was confirmed for this Tess revision; operators also need to review the inherited Inkling model's terms and any external tool or dataset obligations.
The model can execute or propose commands inside coding agents. Deployment should keep permission boundaries, filesystem/network sandboxes, audit logs and explicit confirmation for destructive operations.
Choose this release for research into targeted coding fine-tunes of large sparse open-weight bases. Pin the complete merged checkpoint or separately pin both the base and LoRA, and record all package versions and inference settings.
Evaluate on reproducible real repository tasks with code execution tests, failure analysis, actual GPU memory and cost. The developer's favorable benchmark does not automatically justify promoting Tess itself to Leading.
IMPLEMENTATION & EVALUATION
Source-grounded deployment checklist; no Open Model Weights inference or benchmark was performed.
Mutable model names are not immutable checkpoints.
Record exact 40-character SHA, file inventory, config and runtime.
Unpinned dependencies can change outputs.
Repository labels do not cover every downstream dependency.
Check original model license declarations, extra licenses and required notices.
Unlicensed redistribution, privacy violations or incompatible integrations.
Source size and publisher numbers are not peak runtime requirements.
Record hardware, software versions, memory, latency, inputs and failure modes.
OOM, unsupported operators and unstable behavior.
Publisher measurements are not independent model rankings.
Compare pinned baselines on held-out representative tasks, report negative results.
False attribution of vendor performance to OMW.
Proposed tests, not OMW benchmark results. Read the official model card at the pinned source revision ↗ Inspect OMW field evidence ↗
PRIMARY & INDEPENDENT SOURCES
FAQ
No. It is a coding-focused derivative of Inkling 2 Small; the publisher's 594B and 17B counts describe the inherited logical architecture.
The inspected Safetensors metadata measures stored packed tensor values. It is not equivalent to the publisher's 594B logical parameter count.
Yes. The inspected repository includes merged checkpoint shards. A separate LoRA adapter is offered for compatible base deployments.
No. The 88.2% versus 86.3% SWE-Bench Pro subset result is developer-reported and not independently reproduced by OMW.
The checked configuration declares 1,048,576 tokens, subject to actual serving backend and hardware limits.
No. Apache-2.0 is declared in repository metadata; no separate standard license file was confirmed in the inspected revision.
03 · PRACTICAL TECHNICAL PROFILE
Inspect weight-memory estimates, model lineage and runtime evidence separately from promotional or untested performance claims.
DEPLOYMENT PROFILE
These bars visualize weight-only decimal-GB estimates. They exclude KV cache, activations, runtime overhead, optimizer state and sharding.
LICENSE & USE
Commercial use allowed under apache-2.0 conditions
This section summarizes the checked repository evidence. It is not legal advice.
LINEAGE & PROVENANCE
Open Model Weights does not infer a hidden base model or training corpus when the checked sources do not declare one.
RUNTIME SIGNALS
Declared and model-card-mentioned runtimes are discovery signals. They are not treated as Open Model Weights runtime tests unless explicitly measured.
EVIDENCE TRAIL
This is where the article remains auditable: source links on one side, the registry's observed change history on the other.
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.
04 · EVIDENCE AND UPDATES
Inspect dated source checks and any differences between published claims and the evidence; these checks are not independent model benchmarks.
SOURCE-BASED VISUAL BRIEF · NOTABLE
Three evidence-linked visual summaries: parameter scale, illustrative precision memory and verifiable provenance. No performance ranking is implied.
FRESHNESS & VERIFICATION
A verification check does not pretend the model changed. Open Model Weights updates the verification date after a real primary-source check, while factual-update dates move only when retained source evidence changes.
CONFLICT INTELLIGENCE
05 · SOURCE RECORD
Full repository-level attributes, licensing, format details and provenance are available below without hiding or rewriting the original data.
01 · WEIGHTS
model-00001-of-00024.safetensorsmodel-00002-of-00024.safetensorsmodel-00003-of-00024.safetensorsmodel-00004-of-00024.safetensorsmodel-00005-of-00024.safetensorsmodel-00006-of-00024.safetensorsmodel-00007-of-00024.safetensorsmodel-00008-of-00024.safetensorsmodel-00009-of-00024.safetensorsmodel-00010-of-00024.safetensorsmodel-00011-of-00024.safetensorsmodel-00012-of-00024.safetensorsmodel-00013-of-00024.safetensorsmodel-00014-of-00024.safetensorsmodel-00015-of-00024.safetensorsmodel-00016-of-00024.safetensorsmodel-00017-of-00024.safetensorsmodel-00018-of-00024.safetensorsmodel-00019-of-00024.safetensorsmodel-00020-of-00024.safetensorsmodel-00021-of-00024.safetensorsmodel-00022-of-00024.safetensorsmodel-00023-of-00024.safetensorsmodel-00024-of-00024.safetensorsmodel.safetensors.index.jsonmtp.safetensors02 · LICENSE
Classification describes the checked license terms; it is not legal advice.
03 · HARDWARE
Weight-only decimal-GB estimate from total parameter count; excludes KV cache, activations, runtime overhead, optimizer state and sharding.
04 · FORMATS
“Not observed” means no matching repository artifact/name/dtype signal was found; it does not prove no third-party conversion exists.
05 · LINEAGE
06 · TRAINING ASSETS
07 · RUNTIME SUPPORT
Open Model Weights separates model-repository evidence from external runtime-release evidence. An external support statement does not change the official model repository's weight formats and does not imply an Open Model Weights inference benchmark.
08 · SOURCES
09 · VERIFICATION HISTORY
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.
MODELS FOR AI AGENTS · ROLE EVIDENCE
Specialized model. A supporting component. This checkpoint does not, by category alone, serve as the planning and tool-execution controller.
The curated notable tier is not an agent-readiness rating. Model-specific tool-call success, long-horizon task accuracy and compatible runtime configurations are not independently established here.
Source-linked license: apache-2.0 (allowed commercial-use classification; see original terms).
Context: 1048576 configured tokens.
BF16 weights: 766.15 estimated decimal GB, not runtime VRAM.
Compare agent model candidates → · Agent evaluation method → · Inspect original model repository ↗
These records are separately sourced. They do not replace independent model-plus-runtime tests.
No model-specific agent capability evaluation is recorded for this checkpoint. Missing evidence is unknown, not a failed task.
Read evidence details and benchmarks → · Use the agent model selector →