Skip to main content
BenchLM
Advertise

Model profile

Celeris-1 diffusion language model

CelerisCurrentReleased Jul 22, 2026
Data verified

Celeris-1 is Celeris Labs’ first diffusion language model. It has an 8,192-token context window, costs $2 per million input tokens and $6 per million output tokens, and has one source-verified MMLU-Pro result of 75.9%. It remains unranked until broader public benchmark coverage is available.

Overall Score
Unranked
Arena Elo
Not listed
Eligible category ranks
0of 8
Price (1M tokens)
$2 in / $6 out
API pricing
Speed
Not listed
Context
8K

Evidence coverage

1 of 331 tracked benchmarks is published. 1 is verified and 0 provisional. 1 of 8 categories are measured.

Updated July 24, 2026Methodology
Published / tracked
1 / 331
Verified
1
Provisional
0
Categories with evidence
1 / 8

Evidence by category

  • Agentic0 benchmarks
    Not measured
  • Coding0 benchmarks
    Not measured
  • Reasoning0 benchmarks
    Not measured
  • Knowledge1 benchmark
    Verified
  • Math0 benchmarks
    Not measured
  • Multilingual0 benchmarks
    Not measured
  • Multimodal0 benchmarks
    Not measured
  • Inst. Following0 benchmarks
    Not measured
ProprietaryNon-Reasoning
Confidence:
Low
base

BenchLM is tracking Celeris-1, but this profile is currently excluded from the public leaderboard because it still lacks enough non-generated benchmark coverage to rank safely. Only non-generated public benchmark rows appear below.

Celeris-1 is a proprietary model with a 8K token context window. It processes queries without explicit chain-of-thought reasoning, offering faster response times and lower token usage.

Available through Celeris's OpenAI-compatible API in the United States. The model is intended for short, structured, latency-sensitive tasks; Celeris recommends another model for long-form generation.

This profile currently has 1 of 331 tracked benchmarks. BenchLM only exposes non-generated benchmark rows publicly, so missing categories stay blank until a sourced evaluation is available.

Model research

What Celeris-1 is built for

Celeris-1 is a proprietary diffusion language model released on July 22, 2026. Celeris built it for short requests that sit inside an interactive path, including structured extraction, routing, classification, query rewriting, and repeated agent steps. The API follows the OpenAI chat-completions shape and streams responses by default.

Diffusion is the architectural distinction. Autoregressive models normally commit to one token after another; Celeris says its system combines parallel decoding with sequential processing where language needs it. The model weights and full training specification are not public, so the profile records that description without implying an independent architecture audit.

How to read the speed claims

Celeris reports 1,664 output tokens per second at p50 on 49 successful timed requests from a 50-request workload. It also reports a 158 ms p50 response time alongside the 75.9% MMLU-Pro result. Both tests publish prompt shape, sample count, timing basis, and important limitations.

Comparable runtime data requires one method across providers. Celeris’s figures therefore stay attributed in this profile rather than filling the independent speed card. The blank field is deliberate: a fast provider measurement is useful evidence, but it is not interchangeable with a separate benchmark pipeline.

Price, context, and the practical limit

Pay-as-you-go pricing is $2 per million input tokens and $6 per million output tokens. The context window is 8,192 tokens, API access is currently limited to the United States, and Celeris recommends using another model for long-form generation.

That makes Celeris-1 a focused candidate rather than a default model for every workload. The current evidence covers one knowledge benchmark and does not yet answer how the model performs on coding, tool use, multilingual tasks, or long-context retrieval. A production decision still needs a replay set built from the requests the application will actually send.

Primary Celeris-1 sources

Peer position

Exact provisional scores and ranks for the closest listed peers. A score can appear before a model clears the evidence threshold for a rank, so equal scores can have different rank states.

Range 77.44–83.93

  1. Claude Mythos 5
    Anthropic
    #183.93
    Claude Mythos 5 is #1 with a score of 83.93.
    Compare
  2. Claude Fable 5
    Anthropic
    #283.68
    Claude Fable 5 is #2 with a score of 83.68.
    Compare
  3. GPT-5.6 Sol
    OpenAI
    #381.96
    GPT-5.6 Sol is #3 with a score of 81.96.
    Compare
  4. Kimi K3
    Moonshot AI
    #480.96
    Kimi K3 is #4 with a score of 80.96.
    Compare
  5. Claude Opus 4.8
    Anthropic
    #578.34
    Claude Opus 4.8 is #5 with a score of 78.34.
    Compare
  6. Muse Spark 1.1
    Meta
    #677.44
    Muse Spark 1.1 is #6 with a score of 77.44.
    Compare
  7. Celeris-1Current model
    Celeris
    UnrankedNot measured
    Celeris-1 is Unranked with a score of Not measured.

Category percentile

More

Relative position among models eligible for each sourced category. A higher percentile means a stronger position within that category's ranked cohort; 100 is highest.

No eligible category percentile is available from the published evidence yet.

Category evidence

Scores and ranks appear only where this model has published benchmark evidence. Categories without displayable source records remain not measured.

Category scores, ranks, weighting, benchmark coverage, and evidence status
CategoryScore
AgenticWeight 22%0 benchmarksNot measuredNot measured
CodingWeight 20%0 benchmarksNot measuredNot measured
ReasoningWeight 17%0 benchmarksNot measuredNot measured
KnowledgeRank Not rankedWeight 12%1 benchmarkVerified70.9
MathWeight 5%0 benchmarksNot measuredNot measured
MultilingualWeight 7%0 benchmarksNot measuredNot measured
MultimodalWeight 12%0 benchmarksNot measuredNot measured
Inst. FollowingWeight 5%0 benchmarksNot measuredNot measured

Benchmark Details

Rows below have a displayable published verification record. Each source link and provenance note remains in the page HTML while its category is closed. Source-unverified manual rows and generated rows stay hidden.

Knowledge1 benchmark
MMLU-ProProvider exact

Massive Multitask Language Understanding Professional

75.9%Weighted 30%
Source: Celeris: Celeris-1 benchmark reportProvenance: Celeris reports 75.9% on the full MMLU-Pro test set using five-shot chain-of-thought exemplars, strict answer-format scoring, and a zero reasoning budget.

Frequently Asked Questions

What is Celeris-1?

Celeris-1 is the first proprietary model from Celeris Labs, released July 22, 2026. It uses diffusion-based language generation rather than purely token-by-token autoregressive decoding. The hosted API exposes an 8,192-token context window and is designed for short, structured requests where response time matters.

Is Celeris-1 a diffusion language model?

Yes. Celeris describes Celeris-1 as a diffusion language model that generates multiple tokens per model invocation. Its production system combines parallel decoding with sequential processing where needed. The weights and full architecture are not public, so the page records the provider’s description without treating it as independently reproduced.

How much does the Celeris-1 API cost?

Official pricing is $2 per million input tokens and $6 per million output tokens, metered separately. The pay-as-you-go plan uses the same Celeris-1 model and includes streaming through an OpenAI-compatible API. Enterprise pricing, dedicated clusters, and VPC deployment are quoted separately.

How fast is Celeris-1?

Celeris reports 1,664 output tokens per second at p50 on its 1,000-token workload and 158 ms p50 response time on its MMLU-Pro run. These are provider-run measurements with published methods, not BenchLM’s independent runtime series, so the speed field remains blank until a comparable external snapshot is available.

What is the Celeris-1 MMLU-Pro score?

Celeris reports 75.9% on full MMLU-Pro using five-shot chain-of-thought exemplars, strict answer extraction, and a zero reasoning budget. BenchLM displays that exact sourced row but leaves Celeris-1 unranked because one benchmark does not provide enough category coverage for a stable overall score.

Related Resources

More from Celeris LLM Pricing Comparison BenchLM Methodology The AI Race Which LLM Should I Use? Benchmark Explainers
Last updated: July 24, 2026 · Runtime metrics stay blank until BenchLM has a sourced snapshot.

Choose with this week’s evidence

Join 2,000+ readers for ranking moves, new releases, pricing changes, and the evidence behind them.

Free. One email per week.