The KAGAMI mark КАГАМИ
kagami.bg/academy · lesson · machine-readable viewTESTED 2026-10-11 · VERIFIED 2026-10-01 · UPDATED 2026-10-11
IDENTITY
module
KW I26 · Which model for what: a reference matrix
series
KAGAMI Way · Track I — Infrastructure
level
Intermediate (reference)
duration
~9 min lookup
trust_label
TESTED 2026-10-11 (six internal trial passes 2026-09-14 to 2026-09-19, plus three document-extraction rounds 2026-10-10 to 2026-10-11) · VERIFIED 2026-10-01 (model names, tags and licences re-checked at the public sources) · UPDATED 2026-10-11
state_as_of
2026-10-11 — our own observations on own hardware, NOT a benchmark
language
human view: en · bulgarian edition: /academy/moduli/KW_I26_Model_Verdict_Matrix.html
next
KW_I27_Product_Layer_EU_Boundary.html · The product layer
PURPOSE

Quick lookup of which model we use for which job, as of 2026-10-01. Local models (Ollama) and cloud models (Claude) are listed with a verdict: use / niche / avoid. Scores come from small internal trials (6 guard cases, 3 compliance questions, 10 vertical tasks, 5 retrieval queries); treat them as our experience on our own hardware, not as a leaderboard.

KEY CONCEPTS
COMMANDS / PATHS
CHECKLIST
NEXT MODULE

KW_I27_Product_Layer_EU_Boundary.html · The product layer: where the assistant lives and where the data goes · offer: Quick experiment (kagami.bg/stalbata/)

SOURCES
TAGS
model-selectionreferencelocal-llmollamaembeddingvisionclaude-lineupthinking-controldocument-extractionnot-a-benchmark
TESTED · 11.10.2026 VERIFIED · 01.10.2026 UPDATED · 11.10.2026

Which model for what — a reference matrix

One table to come back to when you choose a model for a task. A reference as of 11.10.2026 — our own observations from trials on our own machine, not a benchmark. New on 11.10.2026: a trial on real accounting documents and the lesson that how you call a model matters as much as which model you pick. The verdict is one of three: use, niche or avoid. A companion to the lessons on the trial method (I20) and on reading documents from a photo (I19).

⏱ ~9 min lookup Reference Local AI our trials · as of 11.10.2026
Local models (Ollama)🔒 local Claude🌐 global model

01What you will learn

02Before you start

03Steps

  1. How to read the matrix

    ✅ use — proven for that role in our trials; ◐ niche — fits one specific case, with a caveat; ⛔ avoid — not for structured output or data. Models carry the exact names from the public catalogues, checked on 01.10.2026.

    📅
    Shelf life: 01.10.2026
    Models change within weeks. This is a reference as of that date and our own experience, not an independent assessment. Before deciding on work with personal data or on a quote, run your own small trial and write the date down.
  2. Text models — the workers 🔒 local

    Six trial passes between 14 and 19.09.2026. The key rules: the universal worker is small; the biggest are not necessarily better; reasoning models need a forced format.

    ModelOur experienceVerdict
    gemma3:12b~25 tps · guard 6/6 · compliance 3/3 · dates ✓ · nomenclature 6/6✅ The universal worker — reader, guard, compliance, dispatcher
    qwen3:14b~22 tps · guard 5/6 · compliance 3/3 · dates ✓ · nomenclature 6/6✅ Second universal — an alternative to 12b
    qwen3:32b~11 tps · vertical 10/10 · guard 6/6, no false alarms · compliance 3/3+3/3 · 0 invalid JSON✅ Premium guard/compliance — dense model, slow
    nemotron-3-super (87 GB in Ollama)~22 tps · vertical 10/10 · guard 6/6 · compliance 3/3+3/3 · runs with context 8k + flash attention + KV q8_0 · answers Bulgarian prompts in Russian✅ with a caveat — perfect on the test but weak for Bulgarian; the official language list does not include Bulgarian
    Mistral-Small-4-119B (Apache-2.0)~50 tps · guard 6/6 · compliance 3/3+3/3 · invoice + EU ✓ · on our setup it loaded with Ollama 0.34.2; the older 0.23.2 did not recognise the architecture✅ Premium compliance/guard — about 43 GB, for high-stakes work, not daily. Weights on Hugging Face; there is no official tag in the Ollama library (checked on 01.10.2026)
    deepseek-r1:32b~10 tps · guard 6/6 · compliance 3/3 — the strongest of the smaller ones◐ Top compliance, but slow; raw, it spills its reasoning
    qwen2.5-coder:7b~44 tps · sums ✓ · guard 0/6 and 6 false alarms · date 2023 instead of 2026◐ Only calculations and code — not for judgment
    mistral:7b~44 tps · extraction ✓ · guard 5/6, but 3 false alarms◐ Extraction ok; the guard is noisy
    gemma3:4b~73 tps · fast, but compliance 0/3 · dates ✗◐ Light routing only — not for data
    qwen3:30b-a3b~78 tps · raw: 0/10 valid JSON (reasoning leaks) · with format=json: 0 invalid out of 10 · guard 6/6 · in free text: 3 invented refusals◐ Tamed with format=json — fine for extraction, not for refusals or free text
    Qwen2.5-7B-Instruct-1M (Apache-2.0)~47 tps · vertical 8/10 · guard 4/6 (missed 2 leaks) · with unbounded context it froze the machine; we capped it at 8k◐ Not for a guard — long context only with a cap. Weights on Hugging Face; there is no official tag in the Ollama library (checked on 01.10.2026)
    gemma3:12b-it-q8_0 (13 GB)Q8 variant of 12b; not in the three-pass test◐ More precise, heavier — Q4 is enough for everyday work (our choice, not a measurement)
    gemma3:27b~12 tps · guard 6/6 · nomenclature 1/6 · slow⛔ No gain over 12b for structured data
    qwen3:8b~39 tps · guard 0/6 (missed everything)⛔ Guard-blind — dangerous for judgment
    llama3.1:8b~41 tps · wrong total, wrong dates, guard broke the JSON⛔ Silent errors — not for data
    gemma4:26b-a4b (q8 and q4)11.10.2026, document extraction: without an explicit think: false — 32–36% and up to 10 failures; with it — q8: 64%, 0 failures, median ~7 s✅ A reliable second opinion for extraction — only with an explicit think: false
    gpt-oss:20b~58 tps · raw: 0/10 valid JSON (reasoning leaks) · 11.10.2026: via /api/chat with think: "low" — 55% on extraction (from 14%)◐ Only when called right — /api/chat + think: "low"; still not for Bulgarian (see the addendum of 23.09)
    deepseek-r1:70b~5 tps · raw: 0/10 · very slow · 11.10.2026: on extraction it timed out on most documents (~2 min each)⛔ Not for extraction; heavy reasoning, 32b is enough
    ✅
    A big model is not the problem — unbounded context is
    The biggest model in the table passed the whole battery once the context was capped and the cache quantized. The earlier crash came from a model with a context of up to 1 million tokens that ate the memory until the machine froze. Cap the context, don't fear the size.
  3. Search models — Bulgarian 🔒 local

    A trial of 5 Bulgarian queries over our own documents. The criterion is which document comes first and how widely the right ones separate from the wrong ones.

    ModelOur experienceVerdict
    snowflake-arctic-embed25/5 · all in first place · widest separation (0.08–0.25)✅ Best for Bulgarian search
    bge-m35/5 · all in first place · 0 invalid vectors✅ Excellent, on par
    granite-embedding:278m5/5 (one in third place) · tiny and fast◐ Light fallback
    nomic-embed-text3/5 · blind on 2 Bulgarian queries⛔ Rejected for Bulgarian
  4. Vision models and reading from a photo 🔒 local

    The lesson on reading documents (I19) describes pass 3, when the four open vision models failed on Cyrillic. Pass 4 changed the picture for one of them: the problem with qwen2.5vl:7b was a missing small num_ctx, not the model.

    Model / solutionWhat happenedVerdict
    OCR (Tesseract, Bulgarian) → gemma3:12bCorrectly extracted the total and all 7 line items from an invoice✅ This works — reading from a photo
    qwen2.5vl:7bWith a small num_ctx it read an electricity-meter value correctly ("018565,7") in ~4 s; the pass 3 failure (timeout) came from the missing context limit✅ OCR works (small context)
    minicpm-v:8bGarbled text ("СУМА" became "КВМА")⛔ Fails on Cyrillic
    granite3.2-vision:2bMisread the meter digits⛔ Fails on Cyrillic
    llava:34bRefused: "I don't see an image"⛔ Fails on Cyrillic

    A scene narrator is not OCR. For describing a picture (not reading text), our trials on 16.09.2026 with a small num_ctx gave good results:

    ModelOur experienceVerdict
    qwen2.5vl:7bDescribes scenes vividly and correctly in Bulgarian (scene, not text)✅ Fast narrator by default
    gemma3:12bWarm, accurate Bulgarian narration✅ Narrator with richer language
    gemma3:27bPoetic and emotional, but slow and heats the machine (~60 °C under our load)◐ The special-occasion narrator — not for the marathon
  5. The Claude cloud models 🌐 global model

    The current lineup per Anthropic's official page, checked on 01.10.2026. Prices are not in this reference — see the official page; if you write a quote, calculate in euro with VAT on the day of the quote. The "How we use them" column is our experience, not the vendor's recommendation.

    ModelOfficial descriptionHow we use them
    Claude Fable 5.1For demanding reasoning and long-horizon agentic work · context 1M · output up to 128KThe hardest work: architecture documents and the final check
    Claude Opus 5.5For long-running agentic coding and knowledge work · 1M · 128KThe main "thinking" model in working sessions
    Claude Sonnet 5.5The best combination of speed and intelligence · 1M · 128KExecuting a clearly given address and task
    Claude Haiku 4.5The fastest model with near-frontier intelligence · 200K · 64KChecks and routine
    🧭
    Our delegation rule
    The smallest suitable model for routine; the stronger one only where thinking happens. Anything irreversible waits for a human. This arrangement is our own choice, not a vendor ranking.
  6. Three settings that saved the big models 🔒 local

    You cap the context, quantize the cache and force the format. The two variables are set on the Ollama server; read them in the official FAQ before touching them.

    server environment · less memory for the cache
    OLLAMA_FLASH_ATTENTION=1
    OLLAMA_KV_CACHE_TYPE=q8_0
    request body · context cap, no keep-alive, forced JSON
    {
      "model": "<model>",
      "prompt": "<task>",
      "stream": false,
      "format": "json",
      "keep_alive": 0,
      "options": { "num_ctx": 8192 }
    }
    ⚠️
    Rule 00
    One heavy model at a time, keep_alive: 0, load at or below 80%. A forgotten service once caused an out-of-memory crash — before a heavy task, check what else is hanging around.
    🏆
    The two golden rules
    Definitions beat the model — a clean vocabulary in the prompt lifts accuracy more than a pricier model. An imposed schema rescues — even a weak model returns valid JSON when the format is set. The full method is in I20.
  7. Addendum of 23.09.2026 — discipline in Bulgarian

    In a council run in Bulgarian (one task, strict rules) gpt-oss:20b invented numbers and drifted into Russian — don't use it for tasks in Bulgarian. gemma3:27b invented nothing, but added nothing new either. gemma4:26b is a candidate for a fast classifier — not yet a verdict. No local model proved fit to be a critic or a judge (our experience, a small sample; details in the lesson on the council).

    🗣️
    An adjacent finding: Voxtral TTS
    Voxtral-4B-TTS-2603 from Mistral has open weights, voice cloning and 9 languages — Bulgarian is not on the list. The licence is CC-BY-NC-4.0 — non-commercial use only. As of 19.09.2026 the self-hosting stack had no image for our ARM machine, so for now we use it through the provider's API (our choice).
    📄
    An adjacent finding: Mistral OCR 4
    A strong model for documents, but as of 19.09.2026 self-hosting it was offered only to enterprise customers on request — so it was dropped. The local path (qwen2.5vl:7b with a small num_ctx, or OCR → text model) stays. ⚠️ The provider's terms were not re-checked on 01.10.2026.
  8. Addendum of 10–11.10.2026 — extraction from accounting documents 🔒 local

    We tried 11 models on 22 real accounting documents of one company — utility and telecom invoices, an insurance document, online-payment receipts and the company's own outgoing invoices. The documents were processed only in our machine's memory and deleted after every round.

    📏
    What we measure — and what we don't
    The number is the share of documents whose extracted fields pass the deterministic checks: the company-ID checksum, the arithmetic "base × rate = VAT" and the dates. It is not field accuracy — for the incoming documents we had no ready answer key. It says "the numbers are internally consistent", not "everything was read correctly".
    Model · how it is calledPassed the checksFailedMedian per document
    qwen3:14b · /api/generate68%12.5 s
    gemma3:12b · /api/generate73%130 s
    gemma4:26b-a4b q8 · think: false64% (without it: 32%)07 s
    mistral:7b59%19 s
    gpt-oss:20b · /api/chat · think: "low"55% (the old way: 14%)27 s
    Mistral-Small-4-119B50%210 s
    qwen3:14b · /api/chat · think: false41%546 s
    qwen3:32b · gemma3:27b59% · 32%1–625–36 s
    Full thinking (think: true, qwen3)27–41%7–1241–113 s
    deepseek-r1:70b · think: truetimes out on most documents (~2 min each)
    ✅
    How you call a model matters as much as which model you pick
    gemma4 has a "thinking" capability, and if think is not set explicitly it thinks silently and returns nothing — an explicit think: false doubled the result and removed every failure. gpt-oss wants think: "low" as a string (not false — that breaks its template) and structured output only through /api/chat. For qwen3 it is the opposite: /api/chat made it worse, so it stays on /api/generate. Full thinking did not help extraction and was 5–15 times slower.
    gemma4 · extraction · explicitly without thinking
    POST /api/generate
    {
      "model": "gemma4:26b-a4b-it-q8_0",
      "prompt": "<task + document text>",
      "think": false,
      "format": <JSON schema of the fields>,
      "stream": false,
      "options": { "num_ctx": 8192, "num_predict": 2048, "temperature": 0 }
    }
    gpt-oss · extraction · "low" thinking as a string
    POST /api/chat
    {
      "model": "gpt-oss:20b",
      "messages": [{ "role": "user", "content": "<task + text>" }],
      "think": "low",
      "format": <JSON schema of the fields>,
      "stream": false,
      "options": { "num_ctx": 8192, "num_predict": 900, "temperature": 0 }
    }
    // read message.content; message.thinking is not kept
    🏆
    Rules before the model
    With a known invoice layout the fields are read by their labels ("VAT number", "Tax base", "Amount due") and checked deterministically: 100% correct in 0.13 s per document, without a single model call. The model only fills what is missing. Hence the pattern we use: a fast model first → a second opinion only when the checks fail → a per-supplier template, confirmed by a human, from the second month on.
    🧪
    Real documents catch what tests miss
    Every real-data run in this round found a defect that hundreds of synthetic tests did not: a truncated model answer, a date with slashes instead of dots, a model writing the field name instead of a value. Measure on your own documents — in memory, nothing written to disk — before you trust a number.

    A model with a new architecture may need a newer Ollama release than the one you run. Instead of upgrading the working server, run a second, newer one side by side.

04Check

1. What is this matrix?

2. Which is the universal local worker in our trials?

3. How does a reasoning model (for example gpt-oss:20b) behave "raw" when asked for JSON?

4. What caused the crash with the model that had a context of up to 1 million tokens?

5. What doubled gemma4's result on document extraction?

05What's next

06Sources

  1. The Ollama library — names and tags of the local models: gemma3, gemma4, qwen3, qwen2.5vl, nemotron-3-super, bge-m3, snowflake-arctic-embed2.
  2. Mistral-Small-4-119B-2603 — the weights and the Apache-2.0 licence.
  3. Qwen2.5-7B-Instruct-1M — a long-context model, Apache-2.0 licence.
  4. Voxtral-4B-TTS-2603 — open weights, CC-BY-NC-4.0 licence.
  5. Claude — models overview — the official lineup, context and output limits.
  6. Ollama — FAQ — the cache settings and capping the context.
  7. Ollama — releases — when each architecture was added.
  8. Ollama — thinking — the think field and its levels.
  9. Ollama — structured outputs — format with a JSON schema.
  10. /api/generate and /api/chat — the two request endpoints.
  11. KAGAMI's own trials, 14–19.09.2026: small samples of Bulgarian documents on our own machine. Our experience, not a benchmark.
  12. KAGAMI's own trials, 10–11.10.2026: 22 real accounting documents, 11 models, three rounds, processed in memory only. Our experience, not a benchmark.