Selected Japanese general-purpose research checkpoint: condition 1.2, trained from Nemotron 3 Nano Base using the provider’s Japanese and financial data blend.
No model-specific prompt-caching contract or cached-input price verified. Training caches and recurrent-state dtype settings are not a published prompt-cache API.
Published price
Not collected
No numeric production price verified for this exact model or research checkpoint.
Evaluation configuration only. Output-length experiments at 8192 and 16384 also change sampling to temperature 0.6/top-p 0.95, so those results must remain separate.
Access & pricing
Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.
Provider display label is not a verified API request selector.
Access
Selected checkpoint documented in a research report. No public weight release, commercial API or production rollout verified; the descriptive Atlas name is not an API model ID.
Regions
No production processing region or deployment agreement verified for this exact checkpoint.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.
Caching
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
01 / INITIAL INPUTProcess & retain
The system processes eligible input and retains reusable state.
02 / LATER REQUESTCheck available state
Reuse depends on matching input, session state and the cache policy.
03 / CACHE HITReuse & generate
Reusing state can reduce repeated work. Any charges follow the billing rules below.
ABEJA / CODAS research
Not verified
How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
What the measurements say
Swallow Leaderboard v2 overall
ABEJA · vendor-reported
57.92score / 100
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
September 30, 2026 provider experiment. Swallow evaluation uses a 32,768-token context setting and 16,384-token output cap; financial evaluation uses a 4,096-token output cap. Scores are on a 0–100 scale. EDINET official aggregation excludes unparsable outputs; it is not comparable to all-sample accuracy without the stated denominator.
Provider reanalysis counts unparsable answers as wrong. Three task means over 451 Earnings Forecast, 224 Fraud Detection and 496 Industry Prediction examples; 1,171 total. Context setting 32,768, output cap 4,096. This is not the official parseable-only score.
Lower is better; 1,538 / 18,571 responses across eight Japanese Swallow tasks; MT-Bench excluded. Reasoning-only responses count as missing a final answer.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
Provider identifies this model or selected research checkpoint, but no matching public weights were found in the reviewed official inventory. Closed-weight classification is inferred for this exact artifact. Public base weights, training recipes and other ABEJA releases do not establish public availability of this derivative; private distribution terms remain unverified.
