Best LLM for Summarization in 2026
For summarization, Muse Spark 1.3 Contributor is our pick: $0.10/M tokens on a Document summarization workload, 1.0M context.
Summarization is a long-input, short-output workload — you pay mostly for the input tokens, so a model's input price and context window matter more here than raw output quality.
Quick answer: What is the best LLM for summarization?
Muse Spark 1.3 Contributor, from Meta, is the best fit for summarization at $0.10 per million task tokens on a Document summarization workload, with a 1.0M-token context window. No cheaper value pick beats it for this task.
What evidence supports the Summarization recommendation?
Reproducible Summarization evidence and decision rubric
| Test / run | Prompt and verification | Hard rule |
|---|---|---|
| Code Snippet | exact prompt + 20 recorded runs | Correct iterative algorithm and code-only output |
| Hard Algorithm | exact prompt + 4 recorded runs | 5,000-case harness; O(log n) partition and correct edge cases |
Run dates: budget suite 2026-06-16T20:31:30.728Z; premium suite 2026-06-21T00:00:00.000Z. Results are not a claim about every repository or prompt.
Two-test rubric, failure analysis, and task-shaped ranking
| Model | Accuracy | Latency | Output tokens | Run cost | Failure / qualification note |
|---|---|---|---|---|---|
| GPT-5.4 Pro | 100/100 | 206116 ms | 548 | $0.103 | Passes all 5,000+ randomized cases and every edge case; genuine O(log(min(m,n))) partition search, returns a float, code-only, and explicitly raises on two empty lists. Correct — but the slowest run by far (over three minutes), and now that real usage is reported, comfortably the most expensive. |
| Claude Opus 4.8 | 100/100 | 3790 ms | 382 | $0.011 | Passes all 5,000+ randomized cases and every edge case; genuine O(log(min(m,n))) partition, float return, code-only. The fastest correct solution in this task. |
| GLM 5.2 (Max) | 100/100 | 47713 ms | 2148 | $0.010 | Passes all 5,000+ randomized cases and every edge case with a genuine O(log) partition and float return. The <think> block ahead of the code is GLM's reasoning channel surfaced by our gateway, not reasoning dumped into the answer — GLM's actual content is the clean code block — so it scores level with the other correct solutions, as the cheapest of them. |
| Gemini 3.1 Pro | 99/100 | 17413 ms | 320 | $0.004 | Passes all tests with a clean, minimal O(log) partition and float return, code-only. Docked one point only because two empty lists yield NaN rather than an explicit guard (not required by the prompt). |
Availability caveat: short code tests do not establish repository-scale debugging, multi-file tool use, or agent reliability. The fastest acceptable verdict must therefore clear the correctness rule before speed is considered.
Task-shaped cost ranking (20,000 tasks/month)
| Rank | Model | Effective monthly | Measured verbosity |
|---|---|---|---|
| 1 | Amazon Nova Micro | $18.26 | 0.76× |
| 2 | Amazon Nova Lite | $32.74 | 0.91× |
| 3 | GPT-5 Nano | $36.00 | Unavailable; neutral fallback |
| 4 | Gemini 2.5 Flash Lite | $56.00 | Unavailable; neutral fallback |
| 5 | GPT-OSS 20B | $65.16 | 2.93× |
| 6 | Ministral 8B | $65.58 | 0.93× |
Verified 2026-08-08. full prompt/run evidence →
Try these models for Summarization →Volume, requirement-gate, and pricing cross-check for Summarization
Monthly spend ladder at Document summarization shape
| Calls / month | Overall pick monthly | Budget pick monthly | Overall − budget delta |
|---|---|---|---|
| 5,000 | $25.80 | $25.80 | $0.0000 |
| 10,000 | $51.60 | $51.60 | $0.0000 |
| 20,000 | $103.20 | $103.20 | $0.0000 |
| 50,000 | $258.00 | $258.00 | $0.0000 |
Monthly cost = task price/M × (50,000 input + 800 output tokens) × calls ÷ 1,000,000, at 0.5×, 1×, 2×, 5× the published 10,000-call/month baseline.
Requirement-gate margin for the picked models
| Model | Requirement | Measured value | Margin / result |
|---|---|---|---|
| Muse Spark 1.3 Contributor | 128K required | 1.0M measured | +921K headroom |
| Muse Spark 1.3 Contributor | 128K required | 1.0M measured | +921K headroom |
| Gemini 3.1 Pro | 128K required | 2M measured | +1.9M headroom |
Requirement gates are hard filters, not down-ranking: a model failing any row here is excluded from Summarization candidates entirely, regardless of price or speed.
Pricing-page cross-link for each pick
| Model | Task-weighted $/M | Monthly at published volume | Measured throughput |
|---|---|---|---|
| Muse Spark 1.3 Contributor | $0.10 | $51.60 | Unavailable |
| GPT-OSS 120B (Cerebras) | $0.36 | $181.00 | 2450 tok/s |
| Gemini 3.1 Pro | $2.16 | $1096.00 | 55 tok/s |
Evidence coverage: 0 of 49 candidates have a graded run. No graded accuracy evidence exists for this task; the ranking above is a requirements-and-price match, not a quality claim.
Test the Summarization picks side by side →Verified 2026-08-08. "Unavailable" means no compatible dated evidence was found for that field; it is never treated as zero. Dated source · Full ranking and rubric
Batch 10 summarization decision depth
1. Compression-ratio planner
| Target output | Output tokens | Total API cost | Input spend | Output spend |
|---|---|---|---|---|
| 0.5% | 250 | $0.0050 | $0.0050 | $0.0001 |
| 1% | 500 | $0.0051 | $0.0050 | $0.0001 |
| 2% | 1,000 | $0.0052 | $0.0050 | $0.0002 |
| 5% | 2,500 | $0.0055 | $0.0050 | $0.0005 |
Output tokens = source tokens × target compression ratio. Total API cost = exact input + output spend from MODEL_PRICING; compression ratios are User-supplied targets.
2. One-pass vs map-reduce
| Mode | Chunk overlap | Calls | Final-synthesis token inputs | Total API cost | Latency / quality |
|---|---|---|---|---|---|
| Direct one-pass comparator | 0% | 1 | 50K | $0.0052 | Direct input + 800 output; latency/fidelity Unavailable |
| Map-reduce (8K chunks) | 1K (10%) | 7 | 6K | Final $0.0007 · total $0.0074 | Latency/fidelity: Unavailable |
| Map-reduce (16K chunks) | 2K (10%) | 4 | 3K | Final $0.0005 · total $0.0075 | Latency/fidelity: Unavailable |
| Map-reduce (32K chunks) | 3K (10%) | 2 | 2K | Final $0.0003 · total $0.0070 | Latency/fidelity: Unavailable |
The direct one-pass row is the explicit comparator. Map-reduce uses 10% overlap, mapCalls = ceil(source ÷ (chunk − overlap)), and accounts for both billed chunk inputs (including overlap) and final synthesis phase tokens.
3. Human-review cost boundary
| Review minutes / document | Hourly rate | 10K-document total | Evidence boundary |
|---|---|---|---|
| 0 | $30/hour User-supplied | $51.60 | Review quality rate: Unavailable |
| 5 | $30/hour User-supplied | $25051.60 | Review quality rate: Unavailable |
| 15 | $30/hour User-supplied | $75051.60 | Review quality rate: Unavailable |
Total = 10,000 × (API cost per document + review minutes × hourly rate ÷ 60). Review minutes and hourly rate are User-supplied. Stale-data behavior: dated registry values are snapshots. If a source is older than the page verification date, or a provider changes its policy/pricing/model, re-verify before production use; unknown values remain Unavailable.
Verified 2026-08-08. Luna is the data owner. “Unavailable” means no compatible dated evidence was found; it is never zero, an estimate, or a guessed policy. Stale-data behavior: dated registry values are snapshots. If a source is older than the page verification date, or a provider changes its policy/pricing/model, re-verify before production use; unknown values remain Unavailable. Dated registry source · Run this scenario yourself →
Batch 59 · server-rendered evidence boards · verified 2026-09-07
Intent answer: High-quality summarization balances factual fidelity (no hallucinations), appropriate compression ratios, key point retention, and strict constraint adherence. Output verbosity controls dictate true economic efficiency. Verified 2026-09-07.
Demand evidence: Qualitative demand: summarization quality benchmarks reviewed 2026-09-07; exact US monthly volume is unavailable.
Scope boundary: Compare AI models for document summarization based on factual consistency, information density, hallucination rates, and verbosity-adjusted cost. Exact joins required; unresolved joins render Unavailable.
Summarization factual consistency & hallucination audit
Deterministic formula / rule: fidelity = verified_factual_claims / total_claims_made * 100; zero-hallucination rubric.
Boundary: Owns summarization accuracy and faithfulness evaluation.
| Frozen scenario / field ID | Exact identity and evidence fields | Result | State |
|---|---|---|---|
batch59-task-summary-m1-r1news article executive briefing | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$news article executive briefing; model ID; factual accuracy %; hallucination incidence %; key entity retention %; constraint compliance %; summarization grade; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 1 rule is reproducible but no production observation is claimed |
batch59-task-summary-m1-r2meeting transcript action items | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$meeting transcript action items; model ID; factual accuracy %; hallucination incidence %; key entity retention %; constraint compliance %; summarization grade; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 1 rule is reproducible but no production observation is claimed |
batch59-task-summary-m1-r3earnings call financial summary | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$earnings call financial summary; model ID; factual accuracy %; hallucination incidence %; key entity retention %; constraint compliance %; summarization grade; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 1 rule is reproducible but no production observation is claimed |
batch59-task-summary-m1-r4technical whitepaper abstract | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$technical whitepaper abstract; model ID; factual accuracy %; hallucination incidence %; key entity retention %; constraint compliance %; summarization grade; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 1 rule is reproducible but no production observation is claimed |
batch59-task-summary-m1-r5dense legal judgment summary | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$dense legal judgment summary; model ID; factual accuracy %; hallucination incidence %; key entity retention %; constraint compliance %; summarization grade; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 1 rule is reproducible but no production observation is claimed |
batch59-task-summary-m1-r6unsupported source document type | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$unsupported source document type; model ID; factual accuracy %; hallucination incidence %; key entity retention %; constraint compliance %; summarization grade; verified=2026-09-07 | Unavailable — exact task-summary evidence join is not closed for "unsupported source document type" | FAIL CLOSED — manual, probe, or source evidence required |
First-party citation: All AI Ask model specifications. Verified 2026-09-07; missing or conflicting joins fail closed.
Compression ratio & verbosity cost efficiency receipt
Deterministic formula / rule: effective_summary_cost = (input_tokens * in_rate) + (output_tokens * out_rate); length penalty factor.
Boundary: Owns verbosity-adjusted summarization token economics.
| Frozen scenario / field ID | Exact identity and evidence fields | Result | State |
|---|---|---|---|
batch59-task-summary-m2-r1ultra-concise 3-bullet summary (50 words) | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$ultra-concise 3-bullet summary (50 words); target word count; actual output tokens generated; verbosity index (vs baseline); cost per 100 summaries; economic efficiency winner; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 2 rule is reproducible but no production observation is claimed |
batch59-task-summary-m2-r21-paragraph executive brief (150 words) | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$1-paragraph executive brief (150 words); target word count; actual output tokens generated; verbosity index (vs baseline); cost per 100 summaries; economic efficiency winner; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 2 rule is reproducible but no production observation is claimed |
batch59-task-summary-m2-r3comprehensive section summary (500 words) | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$comprehensive section summary (500 words); target word count; actual output tokens generated; verbosity index (vs baseline); cost per 100 summaries; economic efficiency winner; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 2 rule is reproducible but no production observation is claimed |
batch59-task-summary-m2-r4detailed analytical digest (1200 words) | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$detailed analytical digest (1200 words); target word count; actual output tokens generated; verbosity index (vs baseline); cost per 100 summaries; economic efficiency winner; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 2 rule is reproducible but no production observation is claimed |
batch59-task-summary-m2-r5uncontrolled verbose model drift | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$uncontrolled verbose model drift; target word count; actual output tokens generated; verbosity index (vs baseline); cost per 100 summaries; economic efficiency winner; verified=2026-09-07 | Unavailable — exact task-summary evidence join is not closed for "uncontrolled verbose model drift" | FAIL CLOSED — manual, probe, or source evidence required |
batch59-task-summary-m2-r6unsupported length parameter | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$unsupported length parameter; target word count; actual output tokens generated; verbosity index (vs baseline); cost per 100 summaries; economic efficiency winner; verified=2026-09-07 | Unavailable — exact task-summary evidence join is not closed for "unsupported length parameter" | FAIL CLOSED — manual, probe, or source evidence required |
First-party citation: All AI Ask pricing registry and formulas. Verified 2026-09-07; missing or conflicting joins fail closed.
Domain-specific summary constraint enforcement board
Deterministic formula / rule: constraint_pass = tone_adherence && length_compliance && no_extraneous_preamble.
Boundary: Owns prompt constraint adherence for summarization pipelines.
| Frozen scenario / field ID | Exact identity and evidence fields | Result | State |
|---|---|---|---|
batch59-task-summary-m3-r1strict "no intro/outro conversational filler" | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$strict "no intro/outro conversational filler"; prompt constraint; candidate model compliance rate; common failure mode; automated sanitization rule; deployment readiness verdict; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 3 rule is reproducible but no production observation is claimed |
batch59-task-summary-m3-r2strict length limit: exactly 3 sentences | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$strict length limit: exactly 3 sentences; prompt constraint; candidate model compliance rate; common failure mode; automated sanitization rule; deployment readiness verdict; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 3 rule is reproducible but no production observation is claimed |
batch59-task-summary-m3-r3reading level target: 8th grade readability | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$reading level target: 8th grade readability; prompt constraint; candidate model compliance rate; common failure mode; automated sanitization rule; deployment readiness verdict; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 3 rule is reproducible but no production observation is claimed |
batch59-task-summary-m3-r4structured format: markdown table summary | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$structured format: markdown table summary; prompt constraint; candidate model compliance rate; common failure mode; automated sanitization rule; deployment readiness verdict; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 3 rule is reproducible but no production observation is claimed |
batch59-task-summary-m3-r5redacted confidential entities summary | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$redacted confidential entities summary; prompt constraint; candidate model compliance rate; common failure mode; automated sanitization rule; deployment readiness verdict; verified=2026-09-07 | Unavailable — frozen task-summary fixture requires an exact source, identity, and result receipt | UNTESTED — module 3 rule is reproducible but no production observation is claimed |
batch59-task-summary-m3-r6unsupported formatting constraint | route=$/best-llm-for/summarization; owner=$task-summary; scenario=$unsupported formatting constraint; prompt constraint; candidate model compliance rate; common failure mode; automated sanitization rule; deployment readiness verdict; verified=2026-09-07 | Unavailable — exact task-summary evidence join is not closed for "unsupported formatting constraint" | FAIL CLOSED — manual, probe, or source evidence required |
First-party citation: All AI Ask measured speed dataset. Verified 2026-09-07; missing or conflicting joins fail closed.
Method and limitations: this board exposes deterministic rules, first-party citations, and dated evidence identities. It does not invent volume, coverage, entitlement, retention, residency, quota, capacity, feature support, quality, price, reliability, or legal conclusions. Run the task-summary evidence flow →
Which models rank highest for Summarization?
"Fit" is a requirements match, not a quality benchmark — it combines price, measured speed, context window, and (where we have run it) graded accuracy on this task. Formula below.
| # | Model | Provider | Fit | Evidence | Task price/M | Tokens/sec | Context | Scored on |
|---|---|---|---|---|---|---|---|---|
| 1 | Muse Spark 1.3 Contributor | Meta | 79 | — | $0.10 | — | 1.0M | price, context |
| 2 | Gemini 2.5 Flash LiteLegacy | 78 | — | $0.10 | — | 1M | price, context | |
| 3 | Gemini 2.5 FlashLegacy | 70 | — | $0.33 | — | 1M | price, context | |
| 4 | Gemini 3.7 Flash | 64 | — | $0.80 | — | 1.0M | price, context | |
| 5 | Gemini 3.5 Flash Lite | 63 | — | $0.33 | 162 | 1M | price, context, speed | |
| 6 | Gemini 3.1 Pro | 63 | — | $2.16 | 55 | 2M | price, context, speed | |
| 7 | DeepSeek V4 Flash | DeepSeek | 61 | — | $0.45 | 132 | 1M | price, context, speed |
| 8 | Muse Spark 1.3 | Meta | 61 | — | $1.30 | — | 1.0M | price, context |
What will Summarization cost?
At 10,000 document summarization calls/month:
| Model | Task price/M | Est. monthly cost |
|---|---|---|
| Muse Spark 1.3 Contributor | $0.10 | $51.60 |
| Gemini 2.5 Flash Lite | $0.10 | $53.20 |
| Gemini 2.5 Flash | $0.33 | $170.00 |
How is the best LLM for Summarization ranked?
Weights: evidence 0%, price 40%, speed 10%, context 50%.
Requirements: ≥128K context. 49 models eligible.
Price and context sub-scores are min-max normalised (log-scaled) within this task's eligible set only. Speed uses measured tokens/sec only — estimated rows are excluded. A model missing a measurement is never scored as zero: its weight is redistributed across the components we do have, and "Scored on" in the table above shows exactly which ones.
Prices verified 2026-08-08.
What related resources help with Summarization?
Where can you find evidence and costs for Summarization?
What are common questions about the best LLM for Summarization?
Why does price dominate this ranking?
Summarization workloads are input-heavy by nature — with a 50K-token document and an 800-token summary, input tokens are over 98% of the bill.
Does a cheaper model summarize worse?
We have not graded this task directly — run your own documents through the top picks below and compare before committing to one at volume.
Should I chunk long documents instead of using a big context window?
Chunking adds complexity and can lose cross-section context; a single large-context call is simpler when the model supports it and the price difference is small.