← All jobs
Edge / WASM-adjacent small-model inference
Run small LLMs/embedders next to users on edge runtimes for privacy and TTFT — not full frontier chat.
At a glance — Oct 2026
Updated 2026-10-27 · forward this page: /jtbds/wasm-edge-small-models#briefScores as of 2026-09-14 (43 days old; Depot marks scores stale after 30 days). The pick is still Depot's current editorial pick; it has not been re-reviewed since. Prices below carry their own dates.
- Depot Pick
- Cloudflare Workers AI · 7.80/10 for this job (curated estimate, high confidence)
Distinct from `edge-llm-inference` emphasis: small/on-device-adjacent catalog + WASM/isolates. Confidence: high for CF Workers AI neuron table freshness.
- Runner-up
- Groq · 8.07/10
Scores higher than the pick on: Capability fit, Price / performance, Developer & agent DX.
- When not to use the pick
- Need LPU-class latency on larger open models (Groq), or Workers AI frontier SKUs require Workers Paid you will not buy.
- Lock-in / migration
- The pick has lower stored lock-in risk than runner-up Groq (4.5 vs 5). Pick exit signals: OpenAI-compatible API (provider tag `openai-compat`). Self-host fallback listed for this job: Gemma models under the Gemma Terms of Use.
- Sourced price (usd_per_million_input_tokens)
Pick's lowest sourced SKU: @cf/ibm-granite/granite-4.0-h-micro $0.017 (usd_per_million_input_tokens) — source, as of 2026-10-03, re-checked 2026-10-27
Cheapest matching SKU: Cloudflare Workers AI @cf/ibm-granite/granite-4.0-h-micro $0.017 — source, as of 2026-10-03, re-checked 2026-10-27. The pick is the cheapest.
Price history: first seen 2026-10-03 · last changed no change recorded · history JSON
- Free option
- Gemma models under the Gemma Terms of Use (free) — source, as of 2026-10-26
- Failure modes to watch
- Runaway spend: Set a budget and a tool-call cap per task. Alert when spend jumps. Stop the run at the cap instead of summarizing a partial trace as success.
- Context overflow: If the input exceeds the window, say so and retrieve a narrower slice. Do not answer from a truncated head as if it were the whole source.
- Lost in the middle: Ask the agent to name the passage it used. Spot-check questions whose only support sits in the middle of the context, not only at the edges.
- Non-determinism: Run the task more than once and compare the end state, not a single transcript. Record the trial count next to any success rate you cite.
- Schema drift: Validate every tool call and every loaded table against the current schema. Quarantine mismatches instead of coercing them.
- Silent truncation: Tools must return whether the payload was truncated and how long the original was. Fail the task if the answer depends on the missing span.
Same-job, same-unit comparisons: Cloudflare Workers AI vs Groq
Cite: Agent Provider Depot, Depot Pick for wasm-edge-small-models: cloudflare-workers-ai, scores as of 2026-09-14. Agents: GET /api/decide?jtbd=wasm-edge-small-models
Free Edge small models options · All sourced prices, cheapest first · for local-offline agents
Depot Pick
Cloudflare Workers AI
Distinct from `edge-llm-inference` emphasis: small/on-device-adjacent catalog + WASM/isolates. Confidence: high for CF Workers AI neuron table freshness.
Pricing (directional): As of 2026-10-24 (developers.cloudflare.com/workers-ai/platform/pricing, fetched 2026-10-24): '$0.011 per 1,000 Neurons' with '10,000 Neurons per day at no charge' on both Workers Free and Workers Paid ('All limits reset daily at 00:00 UTC.…
Cite: Depot Pick for wasm-edge-small-models: cloudflare-workers-ai
Overall
7.8curated_estimate
confidence: high · as of 2026-09-14
Capability fit
6.8
Price / performance
8.5
Latency / reliability
8.2
Developer & agent DX
8.4
Privacy / data retention
8.0
Lock-in risk (higher = worse)
4.5
Maturity / track record
7.8
Lock-in risk: higher = more lock-in (worse). Overall already inverts this dimension.
Sources: Workers AI pricing: https://developers.cloudflare.com/workers-ai/platform/pricing/ · Workers AI: https://developers.cloudflare.com/workers-ai/ · Cloudflare Workers AI product: https://www.cloudflare.com/products/workers-ai/
Best when your agent already runs on Workers and you want colocated small/open models. Not a frontier brain.
Quote for Edge / WASM-adjacent small-model inference
Depot Pick stays Cloudflare Workers AI. Prices below are sourced SKU rows only (as of 2026-10-27). Input and output are not blended. GET /api/quote?jtbd=wasm-edge-small-models
Pick is the cheapest sourced input $/1M among candidates (0.017).
- @cf/ai4bharat/indictrans2-en-indic-1B usd_per_million_input_tokens $0.342 source as of 2026-10-03
- @cf/aisingapore/gemma-sea-lion-v4-27b-it usd_per_million_input_tokens $0.351 source as of 2026-10-03
- @cf/baai/bge-base-en-v1.5 usd_per_million_input_tokens $0.067 source as of 2026-10-03
- @cf/baai/bge-large-en-v1.5 usd_per_million_input_tokens $0.204 source as of 2026-10-03
- @cf/baai/bge-m3 usd_per_million_input_tokens $0.012 source as of 2026-10-03
- @cf/baai/bge-reranker-base usd_per_million_input_tokens $0.003 source as of 2026-10-03
- @cf/baai/bge-small-en-v1.5 usd_per_million_input_tokens $0.02 source as of 2026-10-03
- @cf/cloudflare/clef-flash usd_per_million_input_tokens $0.038 source as of 2026-10-19
- @cf/cloudflare/clef usd_per_million_input_tokens $0.24 source as of 2026-10-03
- @cf/deepseek-ai/deepseek-r1-distill-qwen-32b usd_per_million_input_tokens $0.497 source as of 2026-10-03
- @cf/deepseek-ai/deepseek-v4-flash-0731 usd_per_million_input_tokens $0.44 source as of 2026-10-03
- @cf/deepseek-ai/deepseek-v4-pro-0813 usd_per_million_input_tokens $1.32 source as of 2026-10-03
- @cf/google/gemma-3-12b-it usd_per_million_input_tokens $0.345 source as of 2026-10-03
- @cf/google/gemma-4-26b-a4b-it usd_per_million_input_tokens $0.1 source as of 2026-10-03
- @cf/huggingface/distilbert-sst-2-int8 usd_per_million_input_tokens $0.026 source as of 2026-10-03
- @cf/ibm-granite/granite-4.0-h-micro usd_per_million_input_tokens $0.017 source as of 2026-10-03
- @cf/meta/llama-2-7b-chat-fp16 usd_per_million_input_tokens $0.556 source as of 2026-10-03
- @cf/meta/llama-3.1-70b-instruct-fp8-fast usd_per_million_input_tokens $0.293 source as of 2026-10-03
- @cf/meta/llama-3.1-8b-instruct-awq usd_per_million_input_tokens $0.123 source as of 2026-10-03
- @cf/meta/llama-3.1-8b-instruct-fp8-fast usd_per_million_input_tokens $0.045 source as of 2026-10-03
- @cf/meta/llama-3.1-8b-instruct-fp8 usd_per_million_input_tokens $0.152 source as of 2026-10-03
- @cf/meta/llama-3.1-8b-instruct usd_per_million_input_tokens $0.282 source as of 2026-10-03
- @cf/meta/llama-3.2-11b-vision-instruct usd_per_million_input_tokens $0.049 source as of 2026-10-03
- @cf/meta/llama-3.2-1b-instruct usd_per_million_input_tokens $0.027 source as of 2026-10-03
- @cf/meta/llama-3.2-3b-instruct usd_per_million_input_tokens $0.051 source as of 2026-10-03
- @cf/meta/llama-3.3-70b-instruct-fp8-fast usd_per_million_input_tokens $0.293 source as of 2026-10-03
- @cf/meta/llama-3-8b-instruct-awq usd_per_million_input_tokens $0.123 source as of 2026-10-03
- @cf/meta/llama-3-8b-instruct usd_per_million_input_tokens $0.282 source as of 2026-10-03
- @cf/meta/llama-4-scout-17b-16e-instruct usd_per_million_input_tokens $0.27 source as of 2026-10-03
- @cf/meta/llama-guard-3-8b usd_per_million_input_tokens $0.484 source as of 2026-10-03
- @cf/meta/m2m100-1.2b usd_per_million_input_tokens $0.342 source as of 2026-10-03
- @cf/mistral/mistral-7b-instruct-v0.1 usd_per_million_input_tokens $0.11 source as of 2026-10-03
- @cf/mistralai/mistral-small-3.1-24b-instruct usd_per_million_input_tokens $0.351 source as of 2026-10-03
- @cf/moondream/moondream3.1-9B-A2B usd_per_million_input_tokens $0.3 source as of 2026-10-03
- @cf/moonshotai/kimi-k2.5 usd_per_million_input_tokens $0.6 source as of 2026-10-03
- @cf/moonshotai/kimi-k2.6 usd_per_million_input_tokens $0.95 source as of 2026-10-03
- @cf/moonshotai/kimi-k2.7-code usd_per_million_input_tokens $0.95 source as of 2026-10-03
- @cf/nvidia/nemotron-3-120b-a12b usd_per_million_input_tokens $0.5 source as of 2026-10-03
- @cf/openai/gpt-oss-120b usd_per_million_input_tokens $0.35 source as of 2026-10-03
- @cf/openai/gpt-oss-20b usd_per_million_input_tokens $0.2 source as of 2026-10-03
- @cf/pfnet/plamo-embedding-1b usd_per_million_input_tokens $0.019 source as of 2026-10-03
- @cf/qwen/qwen2.5-coder-32b-instruct usd_per_million_input_tokens $0.66 source as of 2026-10-03
- @cf/qwen/qwen3-30b-a3b-fp8 usd_per_million_input_tokens $0.051 source as of 2026-10-03
- @cf/qwen/qwen3.8-27b usd_per_million_input_tokens $0.45 source as of 2026-10-03
- @cf/qwen/qwen3-embedding-0.6b usd_per_million_input_tokens $0.012 source as of 2026-10-03
- @cf/qwen/qwq-32b usd_per_million_input_tokens $0.66 source as of 2026-10-03
- @cf/zai-org/glm-4.7-flash usd_per_million_input_tokens $0.06 source as of 2026-10-03
- @cf/zai-org/glm-5.2 usd_per_million_input_tokens $1.4 source as of 2026-10-03
- @cf/zai-org/glm-5.3-flash usd_per_million_input_tokens $0.15 source as of 2026-10-03
- @cf/zai-org/glm-5.3 usd_per_million_input_tokens $1.4 source as of 2026-10-03
- @cf/ai4bharat/indictrans2-en-indic-1B usd_per_million_output_tokens $0.342 source as of 2026-10-03
- @cf/aisingapore/gemma-sea-lion-v4-27b-it usd_per_million_output_tokens $0.555 source as of 2026-10-03
- @cf/deepseek-ai/deepseek-r1-distill-qwen-32b usd_per_million_output_tokens $4.881 source as of 2026-10-03
- @cf/deepseek-ai/deepseek-v4-flash-0731 usd_per_million_output_tokens $1.32 source as of 2026-10-03
- @cf/deepseek-ai/deepseek-v4-pro-0813 usd_per_million_output_tokens $3.96 source as of 2026-10-03
- @cf/google/gemma-3-12b-it usd_per_million_output_tokens $0.556 source as of 2026-10-03
- @cf/google/gemma-4-26b-a4b-it usd_per_million_output_tokens $0.3 source as of 2026-10-03
- @cf/ibm-granite/granite-4.0-h-micro usd_per_million_output_tokens $0.112 source as of 2026-10-03
- @cf/meta/llama-2-7b-chat-fp16 usd_per_million_output_tokens $6.667 source as of 2026-10-03
- @cf/meta/llama-3.1-70b-instruct-fp8-fast usd_per_million_output_tokens $2.253 source as of 2026-10-03
- @cf/meta/llama-3.1-8b-instruct-awq usd_per_million_output_tokens $0.266 source as of 2026-10-03
- @cf/meta/llama-3.1-8b-instruct-fp8-fast usd_per_million_output_tokens $0.384 source as of 2026-10-03
- @cf/meta/llama-3.1-8b-instruct-fp8 usd_per_million_output_tokens $0.287 source as of 2026-10-03
- @cf/meta/llama-3.1-8b-instruct usd_per_million_output_tokens $0.827 source as of 2026-10-03
- @cf/meta/llama-3.2-11b-vision-instruct usd_per_million_output_tokens $0.676 source as of 2026-10-03
- @cf/meta/llama-3.2-1b-instruct usd_per_million_output_tokens $0.201 source as of 2026-10-03
- @cf/meta/llama-3.2-3b-instruct usd_per_million_output_tokens $0.335 source as of 2026-10-03
- @cf/meta/llama-3.3-70b-instruct-fp8-fast usd_per_million_output_tokens $2.253 source as of 2026-10-03
- @cf/meta/llama-3-8b-instruct-awq usd_per_million_output_tokens $0.266 source as of 2026-10-03
- @cf/meta/llama-3-8b-instruct usd_per_million_output_tokens $0.827 source as of 2026-10-03
- @cf/meta/llama-4-scout-17b-16e-instruct usd_per_million_output_tokens $0.85 source as of 2026-10-03
- @cf/meta/llama-guard-3-8b usd_per_million_output_tokens $0.03 source as of 2026-10-03
- @cf/meta/m2m100-1.2b usd_per_million_output_tokens $0.342 source as of 2026-10-03
- @cf/mistral/mistral-7b-instruct-v0.1 usd_per_million_output_tokens $0.19 source as of 2026-10-03
- @cf/mistralai/mistral-small-3.1-24b-instruct usd_per_million_output_tokens $0.555 source as of 2026-10-03
- @cf/moondream/moondream3.1-9B-A2B usd_per_million_output_tokens $1 source as of 2026-10-03
- @cf/moonshotai/kimi-k2.5 usd_per_million_output_tokens $3 source as of 2026-10-03
- @cf/moonshotai/kimi-k2.6 usd_per_million_output_tokens $4 source as of 2026-10-03
- @cf/moonshotai/kimi-k2.7-code usd_per_million_output_tokens $4 source as of 2026-10-03
- @cf/nvidia/nemotron-3-120b-a12b usd_per_million_output_tokens $1.5 source as of 2026-10-03
- @cf/openai/gpt-oss-120b usd_per_million_output_tokens $0.75 source as of 2026-10-03
- @cf/openai/gpt-oss-20b usd_per_million_output_tokens $0.3 source as of 2026-10-03
- @cf/qwen/qwen2.5-coder-32b-instruct usd_per_million_output_tokens $1 source as of 2026-10-03
- @cf/qwen/qwen3-30b-a3b-fp8 usd_per_million_output_tokens $0.335 source as of 2026-10-03
- @cf/qwen/qwen3.8-27b usd_per_million_output_tokens $3.2 source as of 2026-10-03
- @cf/qwen/qwq-32b usd_per_million_output_tokens $1 source as of 2026-10-03
- @cf/zai-org/glm-4.7-flash usd_per_million_output_tokens $0.4 source as of 2026-10-03
- @cf/zai-org/glm-5.2 usd_per_million_output_tokens $4.4 source as of 2026-10-03
- @cf/zai-org/glm-5.3-flash usd_per_million_output_tokens $0.5 source as of 2026-10-03
- @cf/zai-org/glm-5.3 usd_per_million_output_tokens $4.4 source as of 2026-10-03
- @cf/black-forest-labs/flux-1-schnell 512x512 tile usd_non_token $0.0000528 source as of 2026-10-03
- @cf/black-forest-labs/flux-1-schnell step usd_non_token $0.0001056 source as of 2026-10-03
- @cf/black-forest-labs/flux-2-dev input 512x512 tile, per step usd_non_token $0.00021 source as of 2026-10-03
- @cf/black-forest-labs/flux-2-dev output 512x512 tile, per step usd_non_token $0.00041 source as of 2026-10-03
- @cf/black-forest-labs/flux-2-klein-4b input 512x512 tile usd_non_token $0.000059 source as of 2026-10-03
- @cf/black-forest-labs/flux-2-klein-4b output 512x512 tile usd_non_token $0.000287 source as of 2026-10-03
- @cf/black-forest-labs/flux-2-klein-9b first MP (1024x1024) usd_non_token $0.015 source as of 2026-10-03
- @cf/black-forest-labs/flux-2-klein-9b input image MP usd_non_token $0.002 source as of 2026-10-03
- @cf/black-forest-labs/flux-2-klein-9b subsequent MP usd_non_token $0.002 source as of 2026-10-03
- @cf/deepgram/aura-1 1k characters input usd_per_1k_characters $0.015 source as of 2026-10-03
- @cf/deepgram/aura-2-en 1k characters input usd_per_1k_characters $0.03 source as of 2026-10-03
- @cf/deepgram/aura-2-es 1k characters input usd_per_1k_characters $0.03 source as of 2026-10-03
- @cf/deepgram/flux audio minute usd_per_minute $0.0077 source as of 2026-10-03
- @cf/deepgram/nova-3 audio minute input usd_per_minute $0.0052 source as of 2026-10-03
- @cf/deepgram/nova-3 (WebSocket) audio minute input usd_per_minute $0.0092 source as of 2026-10-03
- @cf/deepseek-ai/deepseek-v4-flash-0731 usd_per_million_cached_input_tokens $0.014 source as of 2026-10-03
- @cf/deepseek-ai/deepseek-v4-pro-0813 usd_per_million_cached_input_tokens $0.044 source as of 2026-10-03
- @cf/leonardo/lucid-origin 512x512 tile usd_non_token $0.006996 source as of 2026-10-03
- @cf/leonardo/lucid-origin step usd_non_token $0.000132 source as of 2026-10-03
- @cf/leonardo/phoenix-1.0 512x512 tile usd_non_token $0.00583 source as of 2026-10-03
- @cf/leonardo/phoenix-1.0 step usd_non_token $0.00011 source as of 2026-10-03
- @cf/microsoft/resnet-50 usd_per_million_images $2.51 source as of 2026-10-03
- @cf/moonshotai/kimi-k2.5 usd_per_million_cached_input_tokens $0.1 source as of 2026-10-03
- @cf/moonshotai/kimi-k2.6 usd_per_million_cached_input_tokens $0.16 source as of 2026-10-03
- @cf/moonshotai/kimi-k2.7-code usd_per_million_cached_input_tokens $0.19 source as of 2026-10-03
- @cf/myshell-ai/melotts audio minute usd_per_minute $0.0002 source as of 2026-10-03
- @cf/openai/whisper audio minute usd_per_minute $0.0005 source as of 2026-10-03
- @cf/openai/whisper-large-v3-turbo audio minute usd_per_minute $0.0005 source as of 2026-10-03
- @cf/pipecat-ai/smart-turn-v2 audio minute input usd_per_minute $0.00033795 source as of 2026-10-03
- @cf/zai-org/glm-5.2 usd_per_million_cached_input_tokens $0.26 source as of 2026-10-03
- @cf/zai-org/glm-5.3 usd_per_million_cached_input_tokens $0.26 source as of 2026-10-03
- @cf/zai-org/glm-5.3-flash usd_per_million_cached_input_tokens $0.03 source as of 2026-10-03
- Workers Paid neurons usd_per_1k_neurons $0.011 source as of 2026-10-03
0 of 5 candidates have no sourced price row. 2 are excluded from the Input $/1M comparison (chat_completion_input). min sourced input SKU among candidates; not quality-adjusted
Runners-up
Groq
Use when latency is the product. Do not build your whole company on one Groq model SKU — they rotate.
llmlpulatencyopen-models
Fireworks AI
Top pick for production open-model serving with real DX. Prefer over random GPU DIY when you want tokens not…
llmopen-modelsinferencefinetune
Deno Deploy
Nice for TS edge tools. Don't confuse isolate deploy with E2B-class untrusted code sandboxes.
edgetypescriptisolatesdeno
OpenRouter
Best multi-model router for agents that need failover and experimentation. Pin providers in production; don’t…
llmroutermulti-provideropenai-compat
When NOT to use the top pick
Need LPU-class latency on larger open models (Groq), or Workers AI frontier SKUs require Workers Paid you will not buy.
Full ranking (score table)
Ranked with this JTBD's weights. Scores are curated estimates unless marked otherwise. Agents: prefer /api/decide then this table for tie-breaks.
| # | Provider | Overall | Fit | $/perf | Latency | DX | As of |
|---|---|---|---|---|---|---|---|
| 1 | Fireworks AI | 8.4 | 8.6 | 8.8 | 8.4 | 9.0 | 2026-09-14 |
| 2 | Deno Deploy | 8.1 | 7.8 | 8.5 | 8.6 | 9.0 | 2026-09-14 |
| 3 | Groq | 8.1 | 7.8 | 9.0 | 8.2 | 8.8 | 2026-09-14 |
| 4 | OpenRouter | 7.9 | 8.8 | 7.8 | 7.6 | 9.2 | 2026-09-14 |
| 5 | Cloudflare Workers AIDepot Pick | 7.8 | 6.8 | 8.5 | 8.2 | 8.4 | 2026-09-14 |
Score methodology for this job
Distinct from `edge-llm-inference` emphasis: small/on-device-adjacent catalog + WASM/isolates. Confidence: high for CF Workers AI neuron table freshness.
capabilityFit: 18%pricePerformance: 32%latencyReliability: 14%developerDx: 12%privacyPosture: 8%lockInRisk: 8%maturity: 8%