Best Cheap AI Models for Agents: Cost and Failure Modes

The cheapest LLMs that hold up in real agent loops are TokenLab's own catalog entries - gpt-5-nano ($0.05/$0.40 per MTok), gemini-3.1-flash-lite ($0.25/$1.50), gpt-5-mini ($0.25/$2.00), and deepseek-v4-flash ($0.147/$0.294) - because agent workloads burn output tokens fast (tool calls, JSON, chain-of-thought scratch), so output price per MTok matters more than headline input price. This guide compares those against mid-tier and premium models, flags which figures are first-party TokenLab pricing versus third-party numbers that still need independent verification, and explains why some model names you'll see quoted elsewhere don't match what TokenLab lists.

Editor's note: this article no longer claims to provide a latency benchmark. We only discuss latency qualitatively because no controlled TTFT, tokens/sec, or retry-rate dataset was available for these exact model routes at refresh time. Treat the table below as a cost-and-failure-mode shortlist, then benchmark latency in your own agent loop before changing production routing.

Key Takeaways

Source Snapshot

Source	Content Type	Observed Date	Relevance to This Article
PixVerse Platform Docs (docs.platform.pixverse.ai)	Video generation pricing	2026-07-08	Not used for LLM claims - video-only, included for transparency
fal PixVerse V6 model page (fal.ai/pixverse-v6)	Video generation pricing	2026-07-08	Not used for LLM claims - video-only
Black Forest Labs pricing docs (docs.bfl.ml)	Image generation pricing	2026-07-08	Not used for LLM claims - image-only
TokenLab internal model catalog	First-party LLM & media pricing	Current listing	Primary source for all "TokenLab series" prices below
Individual provider pricing pages (OpenAI, Anthropic, Google, DeepSeek, Moonshot, Qwen, MiniMax, Z.ai)	LLM pricing	Not dated in this dataset	Must be independently verified before public budgeting claims

The PixVerse, fal, and BFL sources exist in our research set because they were pulled during a broader media-pricing sweep - they describe per-second video billing and per-image credit costs, which have no bearing on per-token LLM pricing. We're listing them here explicitly so readers can see why they're excluded rather than silently dropping them.

Model And Cost Comparison

All prices below are TokenLab's own first-party catalog listings (per-token, expressed as $/MTok), which is the correct basis for LLM agent cost claims:

Model	Provider	Input $/MTok	Output $/MTok	Agent Fit
gpt-5-nano	OpenAI	$0.05	$0.40	Cheapest option; best for short tool-calling steps, classification, routing
gemini-3.1-flash-lite	Google	$0.25	$1.50	Low-cost, high-throughput steps; verify context window before relying on it for long transcripts
gpt-5-mini	OpenAI	$0.25	$2.00	Balance of reasoning and cost for multi-step planning
deepseek-v4-flash	DeepSeek	$0.147	$0.294	Cheapest output-heavy option in TokenLab's catalog; confirm exact model ID before comparing to external quotes
claude-haiku-4-5	Anthropic	$1.00	$5.00	Stronger instruction-following at moderate cost, good for validation steps
deepseek-v4-pro	DeepSeek	$0.441	$0.882	Heavier reasoning subtasks within the same family
gemini-3.5-flash	Google	$1.50	$9.00	Multimodal agent steps (image + text tool use)
gpt-5	OpenAI	$1.25	$10.00	Fallback for complex planning or ambiguous tool selection
claude-sonnet-5	Anthropic	$2.00	$10.00	Reserve for final verification/QA pass on agent output
gpt-5-pro	OpenAI	$15.00	$120.00	Premium ceiling - rarely justified inside an agent loop

Separately, here are external reference figures supplied without a dated source citation. Treat these as unverified until checked against the provider's own pricing page:

Model	Provider	Input $/MTok	Output $/MTok	Verification Status
deepseek-v4-flash (external listing)	DeepSeek	$0.09	$0.18	⚠️ Differs from TokenLab catalog above - confirm which product/tier this refers to
minimax-m3	MiniMax	$0.30	$1.20	Needs provider confirmation
qwen3.7-plus	Alibaba/Qwen	$0.32	$1.28	Needs provider confirmation
kimi-k2.7-code	Moonshot AI	$0.74	$3.50	Needs provider confirmation
glm-5.2	Z.ai	$0.93	$3.00	Needs provider confirmation
claude-sonnet-5	Anthropic	$2.00	$10.00	Needs provider confirmation
claude-opus-4.8	Anthropic	$5.00	$25.00	Needs provider confirmation
gpt-5.5 Batch/Flex	OpenAI	$2.50	$15.00	Needs provider confirmation; not the standard GPT-5.5 row
gpt-5.5	OpenAI	$5.00	$30.00	Needs provider confirmation
claude-fable-5	Anthropic	$10.00	$50.00	Needs provider confirmation

Implementation Notes

1. Tier your agent by task, not by a single "cheapest" model. Route routing/classification/tool-selection to gpt-5-nano or gemini-3.1-flash-lite; escalate multi-step planning to gpt-5-mini or deepseek-v4-pro; reserve claude-sonnet-5 or gpt-5 for final-answer verification.
2. Cap output tokens aggressively at the cheap tier. Since output cost dominates agent spend, set tight max-token limits on nano/flash-lite calls and only allow longer generations at the escalation tier.
3. Log failure modes per model, not just cost per call. Track truncated JSON, missed tool calls, and hallucinated function arguments separately from token spend - a $0.05/MTok model that fails 15% of tool calls can cost more in retries than a $1/MTok model that fails 2%.
4. Pin exact model IDs in code, never aliases. Given the naming collision shown above (two different "deepseek-v4-flash" price points), hardcode the full vendor+model string and re-verify pricing on each provider update.
5. Re-check external pricing quarterly. Anything without a dated source URL in this article should be re-pulled from the provider's live pricing page before it's used in a customer-facing cost estimate.

When To Use TokenLab

• Single catalog, no per-vendor account juggling - TokenLab exposes OpenAI, Google, Anthropic, and DeepSeek pricing tiers under one API and billing surface, so switching a workflow step from gpt-5-nano to deepseek-v4-flash is a config change, not a new integration.
• First-party, dated pricing you can audit - unlike scattered third-party quotes, TokenLab's catalog numbers used in this article are pulled directly from the platform's own listings, which is why they're the only prices here presented without a "must verify" caveat.
• Built-in tiering for agent pipelines - because TokenLab hosts both the cheap-tier and escalation-tier models side by side, you can A/B cost and failure rate across gpt-5-nano, gemini-3.1-flash-lite, and gpt-5-mini without re-architecting your agent's routing logic.

Related Reading

• AI API pricing comparison
• TokenLab migration guide

FAQ

Why are PixVerse, fal, and BFL cited in this article if it's about LLM pricing? They aren't cited as evidence for any LLM price claim - they're listed only in the Source Snapshot table for transparency about our broader research sweep, and explicitly marked "not used for LLM claims." Every dollar figure in the Model And Cost Comparison table comes from TokenLab's own catalog or is flagged as unverified.

I searched for "deepseek-v4-flash" and found nothing - is this model fake? It's a real entry in TokenLab's internal model catalog, but model names get reused and re-priced across providers and resellers - note that this article itself contains two different price points for a model called "deepseek-v4-flash" ($0.147/$0.294 in TokenLab's catalog vs. $0.09/$0.18 in an external listing). That's exactly why we recommend pinning the exact vendor+model ID in code and re-verifying against the provider's live docs rather than trusting a name alone.

Can I trust the external reference prices (claude-sonnet-5, glm-5.2, etc.) for budgeting? Not yet - those rows are marked "needs provider confirmation" because no dated source URL was available for them at the time of writing. Use the TokenLab catalog figures for planning, and treat the external table as a starting point for your own verification, not a final number.

What's the actual cheapest model for a production agent? By output cost, gpt-5-nano ($0.40/MTok output) is the cheapest in TokenLab's catalog, followed by deepseek-v4-flash ($0.294/MTok output). But "cheapest" should be measured after accounting for retry rate on your specific tool-calling schema - run a failure-mode audit before committing a pipeline to the lowest-cost tier.

Ready to test these models against your own agent workloads with transparent, first-party pricing? Explore the TokenLab model catalog and start comparing costs today.