Pre-retirement tool-use observation: three models through Ollama cloud · V1

Model observation · V1

Sigma5C observed minimax-m2.7, mistral-large-3:675b and nemotron-3-nano:30b through Ollama cloud before the retirement date stated in an emailed vendor notice. The primary window on October 28 and later replicate on October 29, 2026 retained separate execution and timing records. Completion and tool invocation do not establish answer correctness.

Why this observation matters

A lifecycle change creates a finite window for retaining an observation of an existing serving route. A dated record preserves behavior under declared conditions without implying future availability or comparable behavior from another model.

CEO

A historical reference

Preserve a dated view of a retiring serving route before its lifecycle change.

CFO

Evidence for oversight

Keep observed execution, tool use and validation scope distinct when reviewing the record.

CTO

Conditions and traceability

Examine recorded identity and generation conditions before interpreting a trace.

What we did

We reused an earlier bounded tool-use method with a smaller unchanged task subset. Each of three models received the same environment-information reading and computation task subset. We retained model-route identity, timestamps, requests, tool traces, execution states and timing. Generation and tool bounds were reduced and the available tools were restricted; these changes are declared rather than treated as identical conditions to the earlier observation.

Selected observations under these conditions

Recorded execution states

In each window separately, 9 of 9 task records reached the harness completion state. These execution records do not establish correct or satisfactory answers.

Required tool use was recorded

The requested tool class appeared in 9 of 9 task traces in each window separately. This records tool invocation rather than semantic correctness or a general reliability rate.

Two separately recorded windows

The later replicate used the same frozen task subset and recorded conditions. Execution completion and required-tool presence were recorded in both windows; elapsed totals varied. Two windows do not establish response agreement, drift or general reliability.

Selected aggregate charts

Run-total recorded task wall time · primary
minimax-m2.7 · primary
24.7 seconds total
mistral-large-3:675b · primary
15.4 seconds total
nemotron-3-nano:30b · primary
6.8 seconds total

3 task records per model. Totals include inference, tool execution and route delays; they are neither isolated generation latency nor answer-quality scores.

Run-total recorded task wall time · replicate
minimax-m2.7 · replicate
23.2 seconds total
mistral-large-3:675b · replicate
15.4 seconds total
nemotron-3-nano:30b · replicate
6.0 seconds total

3 task records per model. Totals include inference, tool execution and route delays; they are neither isolated generation latency nor answer-quality scores.

Vendor lifecycle context

An Ollama emailed notice dated October 28, 2026 states that these three models will retire from Ollama cloud on November 9, 2026. It names minimax-m3.1-flash as the alternative to minimax-m2.7, mistral-large-4 as the alternative to mistral-large-3:675b, and gemma4:31b as the alternative to nemotron-3-nano:30b. These alternatives are attributed to Ollama and were not observed in this task. The date and mappings are sourced to the emailed notice, not independently confirmed by the inspected public model pages. This notice concerns Ollama cloud availability.

Declared conditions

Observed October 28 and October 29, 2026 UTC through Ollama cloud using a direct API tool-use harness. Each window used the same small task subset covering environment-information reading and computation. Requests were bounded and unseeded, with provider-default thinking behavior, a declared temperature and confined tool execution. The primary and later replicate are complete and shown separately. Route descriptors were recorded, but no immutable remote weight identity was available.

Potential uses

CEO / strategy

Dated transition reference

The records could support a historical review of an existing model-dependent workflow.

CFO / oversight

Evidence scope and oversight

The records could help reviewers distinguish execution completion, tool use and answer validation.

CTO / engineering

Regression-check design

The preserved condition classes and traces could inform a later internal comparison under separately declared conditions.

What's in the data

The retained internal record contains versioned primary and replicate bundles, requested and resolved route identifiers, timestamps, generation settings, full request and response envelopes, tool traces, execution states, transparent bounded checks, aggregate derivations and integrity references. Prompts, responses, task-level records and apparatus details are excluded from this public overview. This description does not assert that the records are currently released through the API.

Provider context

Related delivery-route references are maintained separately from this dated observation. A catalog reference does not transfer these results to another model or route.

Models

Related observations

AI-generated & AI-scored. Every observation on this site is produced and judged by AI, and AI makes mistakes. Treat each result as a recorded observation under a declared protocol, as of its date. Method and marker legend